What’s the vibe around vibe coding?

As OpenAI CEO Sam Altman teases the creative writing skills of an upcoming OpenAI model, many developers and software engineers are embracing something called “vibe coding.” The expression is only six weeks old, but it continues to spark debate on Reddit and various Slack channels—including here at IBM.

Put simply, vibe coding is a term that OpenAI Co-Founder and ex-Tesla AI Senior Director Andrej Karpathy recently used to describe how good LLMs have become at reasoning and coding. (Think: Claude’s Sonnet, or AI code editor Cursor). Now, he wrote on X, developers can “give in to the vibes, embrace exponentials and forget that the code even exists.” As Ars Technica put it: “Instead of being about control and precision, vibe coding is all about surrendering to the flow.”

Joshua Noble, a Technical Strategist at IBM, says he believes that vibe coding started as a joke. But the expression stuck, with incubator Y Combinator even creating a 30-minute video explainer titled “Vibe Coding Is the Future.”

Vibe coding might be the future, but it’s still imperfect today. “I think it's really more like if you were out of ideas, or if you're feeling lazy, just let CoPilot do it for you a little bit and see where it leads you,” Noble says. “I just see stuff, say stuff, run stuff and copy-paste stuff, and it mostly works.”

Ash Minhas, a Manager of Technical Content at IBM, agrees: “Vibe coding is a thing,” he says. “It’s like you can take inspiration and convert it into something.” To illustrate his point, Minhas says he and his brother came up with a prompt to create an app that helps users find their FIRE (Financial Independence, Retire Early) number. “If you have technical skills, it's fantastic,” he says.

Now that AI has arrived in nearly every aspect of the economy, could it disrupt the work of developers and software engineers? The ability to use LLMs to code, debug or test could improve many operations, says Shalini Harkar, a Lead AI Advocate at IBM based in Bangalore.

“These days, developers focus more on solving real-world complex problems that have a high impact in real life, designing efficient architectural designs, enabling faster go-to-market strategies and fostering innovation rather than routine tasks,” she explains.

IBM Distinguished Engineer Michael (Max) Maximilien still asks his team to run manual tests on code. “But increasingly, you don’t need to write the test,” he says. “You can actually just write the code and then ask the LLM to write the test for you, and it does a decent job most of the time. So all of a sudden, you've improved your productivity significantly.”

In addition to this efficiency boost, AI can help less experienced developers to fine-tune their skills and gain hands-on experience at a much faster pace. “You can say, ‘Here's what I'm trying to do,’ and it writes the first version of the code,” Maximilien says. “[Then] you fix it, and you keep moving, right? That's the acceleration.”

At this early stage, vibe coding is still something of a developer in-joke: a form of art reserved for the code connoisseurs. But it may also be a sign that we are ready to redefine what collaboration with AI means in the field of computer science and software engineering.

To some observers, the future of tech might also include more human and machine connections. “The future of coding will prioritize natural language understanding, emotion-aware algorithms and real-time sentiment analysis, making human-computer interactions more intuitive,” says Vrunda Gadesha, a Technical Content Writer at IBM.

And while humans will always be an integral part of the process, AI just makes the coding part a lot easier. “Challenging software engineering will always require a human at some point in the process to look at optimization, or very specific business logic or edge cases,” says Noble. “But when we think of coding—just, ‘I'm going to code up an app that does this’—the amount of work that you can have a large language model do is really increasing. And that is pretty exciting.”