Former Tesla AI director Andrej Karpathy’s “vibe coding” concept, initially a casual hobby, has evolved into a professional standard. Now termed “agentic engineering,” developers orchestrate AI agents, demanding new skills. Industry leaders acknowledge feeling challenged, with AI-generated code increasing significantly, signaling a profound shift in software development.

Andrej Karpathy, the former Tesla AI director who coined “vibe coding” exactly one year ago, says the trend has grown from weekend hobby to industry standard. And 2026, he believes, will push it even further.Marking the anniversary on X, Karpathy admitted his original post was just “a shower of thoughts” he fired off without much consideration. But it struck a nerve. The term now has its own Wikipedia article—longer than the section on his other contributions.“Vibe coding is now mentioned on my Wikipedia as a major memetic contribution,” he wrote, clearly amused.

From ‘accept all’ to agentic engineering

Back in February 2025, Karpathy described vibe coding as almost reckless. He would hit “Accept All” on Cursor suggestions without reading diffs, copy-paste error messages with no comment, and let the code balloon beyond his understanding. It worked for throwaway weekend projects. A custom reading app. A simple webapp. Nothing serious.That has changed. Karpathy now distinguishes between casual vibe coding and what he calls “agentic engineering.” The word “agentic” captures the reality that developers orchestrate AI agents instead of writing code themselves. The word “engineering” signals there is still real skill involved—a learning curve of its own kind.In December, Karpathy posted that he had “never felt this much behind as a programmer.” He called it a “magnitude 9 earthquake” and compared the tools to a “powerful alien tool” handed out with no manual. Sometimes it misfires. Sometimes “a powerful beam of laser erupts and melts your problem.”

Even OpenAI’s CEO Sam Altman feels ‘a little useless’

Karpathy is not the only one feeling the ground shift. Boris Cherny, creator of Claude Code at Anthropic, said he feels behind “most weeks.” New grads may have an edge—they carry no assumptions about what AI can or cannot do.OpenAI CEO Sam Altman offered his own confession after using the company’s Codex app, admitting he felt “a little useless” as the tool outperformed him. Cursor CEO Michael Truell has cautioned against going too far—likening pure vibe coding to building a house without checking under the floorboards.The trend has spread beyond engineers. Google’s Sundar Pichai said vibe coding made development “exciting again.” Microsoft’s Satya Nadella revealed up to 30 percent of the company’s code is now AI-generated. Klarna’s Sebastian Siemiatkowski prototypes in 20 minutes instead of waiting weeks. Shopify’s Tobi Lutke vibe codes medical device software on weekends.

Karpathy’s 2026 outlook for vibe coding

Looking ahead, Karpathy expects gains on two fronts. AI models will keep improving. But he is equally excited about advances in the agent layer—the tools that help developers direct those models.“I feel excited about the product of the two and another year of progress,” he wrote.The goal has shifted. It is no longer about vibing on throwaway projects. The challenge now is claiming leverage from AI without compromising quality. The earthquake continues. And the man who named it is still learning how to hold the alien tool steady.

Read Andrej Karpathy’s full post on vibe coding’s anniversary

A lot of people quote tweeted this as 1 year anniversary of vibe coding. Some retrospective:I’ve had a Twitter account for 17 years now (omg) and I still can’t predict my tweet engagement basically at all. This was a shower of thoughts throwaway tweet that I just fired off without thinking but somehow it minted a fitting name at the right moment for something that a lot of people were feeling at the same time, so here we are: vibe coding is now mentioned on my Wikipedia as a major memetic “contribution” and even its article is longer. lolThe one thing I’d add is that at the time, LLM capability was low enough that you’d mostly use vibe coding for fun throwaway projects, demos and explorations. It was good fun and it almost worked. Today (1 year later), programming via LLM agents is increasingly becoming a default workflow for professionals, except with more oversight and scrutiny. The goal is to claim the leverage from the use of agents but without any compromise on the quality of the software.Many people have tried to come up with a better name for this to differentiate it from vibe coding, personally my current favorite “agentic engineering”:

In 2026, we’re likely to see continued improvements on both the model layer and the new agent layer. I feel excited about the product of the two and another year of progress.