Karpathy Says “Vibe Coding” Is Fading as ‘Agentic Engineering’ Becomes the New AI Coding Era
AI-assisted coding is shifting from casual prompts to structured oversight, with engineers managing intelligent agents instead of writing code directly.
The tech industry’s brief obsession with “vibe coding” may already be giving way to something more structured and powerful. According to AI researcher Andrej Karpathy — the very person who coined the term — the next phase of software development is here, and he calls it “agentic engineering.”
Vibe coding, a phrase that gained popularity over the past year, described a relaxed, experimental way of programming with artificial intelligence. Developers would rely on large language models (LLMs) to generate snippets of code, often for prototypes, demos, or quick explorations. It was fast, creative, and informal — more about intuition than process.
But Karpathy now believes the landscape has changed dramatically.
In a recent post on X, he admitted he never expected the term to spread so widely across the tech world. “Vibe coding is now mentioned on my Wikipedia as a major memetic 'contribution' and even its article is longer,” he noted humorously.
When he first introduced the idea in February 2025, AI systems simply weren’t powerful enough for serious engineering tasks. As he explained, "Large Language Model (LLM) capability was low enough that you'd mostly use vibe coding for fun throwaway projects, demos and explorations."
That limitation no longer exists.
Today’s AI tools are far more capable, allowing developers to depend on them for much larger portions of real-world software creation. Instead of manually writing most of the code, engineers are increasingly delegating tasks to AI systems and supervising the output.
To describe this shift, Karpathy proposed a new term.
“Many people have tried to come up with a better name for this to differentiate it from vibe coding, personally my current favourite 'agentic engineering.’”
The phrase reflects a deeper transformation in how programming work is done. Rather than typing every function line by line, developers now manage multiple AI agents that generate, debug, and refine code. The human role becomes more strategic — setting goals, reviewing results, and guiding decisions.
Karpathy clarified the concept, writing, “'agentic' because the new default is that you are not writing the code directly 99% of the time, you are orchestrating agents who do and acting as oversight.”
However, he cautions that this doesn’t make engineering easier or less skilled. Quite the opposite. Mastering these tools requires new expertise and judgment. As he put it, “'engineering' to emphasise that there is an art & science and expertise to it. It's something you can learn and become better at, with its own depth of a different kind."
His remarks come amid a broader industry shift. NodeJS creator Ryan Dahl recently suggested that human-written code may soon be obsolete, while Google CEO Sundar Pichai has praised AI-driven development. Even OpenAI CEO Sam Altman confessed he “felt sad and useless” after his company released the AI coding app Codex for Mac.
If Karpathy is right, the future of programming won’t be about typing faster — it will be about managing smarter machines.
