Tesla’s former AI chief Andrej Karpathy warns software engineers: 'I’ve never felt…'

Artificial Intelligence

Andrej Karpathy, one of the most influential figures in modern artificial intelligence, has delivered a blunt assessment of where software engineering is heading. In what reads like an open letter to developers worldwide, the former Tesla AI director admitted that he has “never felt this much behind as a programmer,” arguing that the profession itself is being fundamentally rewritten by AI.

Karpathy’s comments carry unusual weight. He previously led Tesla’s Autopilot and AI efforts for five years and is also a cofounder of OpenAI. Yet despite this background, he says the pace of change over the past year has left even seasoned engineers struggling to keep up. According to Karpathy, the role of the programmer is shrinking in a traditional sense, with human-written code becoming increasingly sparse between large blocks generated, modified or orchestrated by AI systems.

At the heart of his warning is the emergence of what he describes as a new “programmable layer of abstraction.” Modern developers, he argues, are no longer just writing functions and classes. They are expected to manage agents, sub-agents, prompts, long and short-term context, memory modes, tool protocols and deep integrations inside modern development environments. This shift demands a new mental model for working with systems that are probabilistic, error-prone and constantly changing, rather than deterministic and predictable.

Karpathy also acknowledged a sense of frustration with himself. He believes he could be significantly more productive if he fully mastered the tools now available, but failing to do so increasingly feels like a personal skills gap rather than a tooling problem. That sentiment has resonated with many engineers who feel the ground shifting beneath them faster than they can adapt.

Interestingly, his remarks also mark a subtle evolution in his own thinking. Earlier in the year, Karpathy popularised the idea of “vibe coding,” a casual approach where developers accept AI-generated changes with minimal scrutiny, especially for side projects. While he once described this as amusing and serviceable for small experiments, his recent experiences suggest clear limits. When building a recent project, he admitted that AI agents were more of a hindrance than a help, forcing him to write most of the code by hand.

Others in the industry echoed both sides of his view. Some developers report working almost entirely through AI systems, barely opening a traditional code editor, while others find that AI still struggles with complex or architectural tasks. Research has also complicated the narrative. Studies have shown that experienced developers can actually become less productive when relying heavily on AI assistants, even though they expect large gains.

Despite these contradictions, Karpathy’s central message is clear. AI is not simply another productivity tool layered on top of existing workflows. It represents a structural change to how software is conceived, written and maintained. Engineers who treat it as a passing phase risk falling behind, while those willing to rebuild their habits and mental models may unlock entirely new levels of leverage.

He closed his remarks with a stark metaphor, describing AI as a powerful alien tool handed to humanity without instructions, even as it triggers a magnitude-nine earthquake across the profession. His advice to developers was simple but urgent: adapt quickly, experiment relentlessly, and roll up your sleeves, because standing still is no longer an option.