Vibe Coding is Set to Disrupt Engineering Careers

The variation in results that AI can produce—from impressively sophisticated to alarmingly flawed—might explain the mixed feelings developers have about the technology. A WIRED survey of programmers conducted in March revealed that 36 percent were enthusiastic about AI coding tools, while 38 percent expressed skepticism.

“There’s no doubt that AI will transform code production,” states Daniel Jackson, a computer scientist at MIT currently investigating how to weave AI into large-scale software development. “However, I wouldn’t be surprised if disappointment follows—perhaps the hype will fade.”

Jackson warns that AI models differ fundamentally from compilers that convert high-level programming languages into more efficient lower-level languages, as they do not consistently adhere to instructions. Sometimes, an AI might perform a task better than the programmer; at other times, it may fail significantly.

He notes that “vibe coding” falters when attempting to create serious software. “There are very few scenarios where ‘mostly works’ is sufficient,” he asserts. “When you care about a piece of software, accuracy is key.”

Software projects often involve considerable complexity, and modifications in one area can lead to issues in another. Experienced programmers understand these complexities, Jackson explains, but “large language models can’t navigate those dependency issues.”

He anticipates that software development may evolve toward more modular codebases with fewer dependencies to address the shortcomings of AI. While AI might displace some developers, Jackson suggests it will also compel many others to rethink their strategies and concentrate more on project design.

Excessive dependence on AI could lead to “a bit of an impending disaster,” Jackson warns, as it may result in “massive amounts of broken code rife with security flaws, alongside a new generation of programmers unable to tackle these vulnerabilities.”

Learn to Code

Even organizations that have already incorporated coding tools into their software development workflows acknowledge that the technology is still too unreliable for broader implementation.

Christine Yen, CEO of Honeycomb, which offers solutions for monitoring large software systems’ performance, remarks that simple or formulaic tasks, such as creating component libraries, are more suited for AI integration. Nonetheless, she states that developers using AI at her company have seen only a 50 percent increase in productivity.

Yen adds that for tasks demanding sound judgment, where performance matters, or where the resulting code interacts with sensitive systems or data, “AI isn’t proficient enough yet to be genuinely beneficial.”

“The challenging aspect of building software systems isn’t just writing lots of code,” she notes. “Engineers will still be needed, at least for now, to provide curation, judgment, guidance, and direction.”

Others indicate that a shift in the workforce may be on the horizon. “We’re not witnessing a decline in demand for developers,” asserts Liad Elidan, CEO of Milestone, a company that assesses the impact of generative AI projects. “We’re seeing less demand for average or underperforming developers.”

“If I’m developing a product, I might have needed 50 engineers, but now I could only require 20 or 30,” says Naveen Rao, VP of AI at Databricks, which aids major businesses in creating their own AI systems. “That is absolutely a reality.”

However, Rao believes that learning to code will continue to be a valuable skill for the foreseeable future. “It’s like saying ‘Don’t teach your child math,’” he explains. Knowing how to optimize computer use will likely remain extremely beneficial.

Yegge and Kim, experienced coders, contend that most developers can adapt to the upcoming changes. In their book focused on vibe coding, they propose new approaches for software development, including modular codebases, ongoing testing, and extensive experimentation. Yegge comments that utilizing AI for software creation is evolving into a unique—if slightly precarious—art form. “It’s about doing this without damaging your hard disk or draining your bank account,” he remarks.