AI is evolving at a rapid pace, reshaping industries, rewriting workflows and even redefining what it means to be a programmer. And now, OpenAI co-founder Andrzej Karpathy has admitted that the change has left even experts struggling to keep up. In a strikingly candid post on He warned that engineers must quickly adapt to a profession that is being “dramatically reshaped” by artificial intelligence, otherwise risk being left behind altogether.

Developers can become 10 times more powerful

Karpathy’s post is a mix of astonishment and urgency, a warning to programmers that the ground beneath their keyboards is rapidly changing. He believes that today’s developers can be “10 times more powerful” if they manage to properly harness the new generation of AI tools, but doing so requires mastering an entirely new “programmable layer of abstraction.”

He listed the dizzying array of concepts now involved in modern software development, “agents, subagents, their pointers, contexts, memory, modes, permissions, tools, plugins, skills, hooks, MCPs, LSPs, slash commands, workflows, IDE integration.” It’s a glimpse of how the once-simple craft of coding has transformed into a vast ecosystem of intelligent, semi-autonomous components, programming and a mix of accelerated engineering, logic and linguistics.

For Karpathy, this isn’t just another technology trend. This is a paradigm shift. Referring to the unpredictable nature of the large language models that power most modern AI systems, he warned, “There is a need to build an omnipresent mental model to account for the strengths and pitfalls of fundamentally stochastic, fallible, incomprehensible, and changing entities.”

This is not the first time Karpathy has raised the alarm

This is not the first time this month that Carpathy has spoken about the increasing complexity of AI systems and their impact on developers. Earlier in December, he quipped that using AI tools like OpenAI’s GPT or Anthropic’s Cloud feels like managing a team of interns with telepathy, which is powerful but often chaotic.

He joked in another post that AI coding assistants are like magic, but magic sometimes sets your desk on fire. The commentary summarized a mix of humor and realism about the unexpected benefits and risks of AI-powered programming.

Carpathy has long been an advocate for developers embracing this new world rather than resisting it. He argues that learning to effectively communicate with AI through signals, context management, and system design is as essential now as learning to code was a decade ago. For someone described as one of the architects of modern deep learning, Karpathy’s humility is astonishing. Even if that feels “behind,” it’s a reminder of how quickly AI is reshaping the industry.