The rise of 'vibe coding'

Until recently, software talent was "the scarcest, most coveted resource" in Silicon Valley, said Semafor. Now, the balance of power is shifting, thanks to a new generation of AI tools that have made it possible for anyone to create apps and websites simply by describing what they want – a phenomenon known as "vibe coding".

These AI tools are able to translate instructions written in plain English into computer code, which can then be used to create software; no need to master programming languages or complex data structures. Already, companies large and small are using these new AI-powered assistants to automate processes that, just a year or two ago, would have taken human software engineers hundreds of hours. Vibe coding is fuelling a new Silicon Valley boom – and, for software engineers, it's an alarming encroachment into their industry.

'Forget that the code even exists'

The term "vibe coding" was coined this year by computer scientist and OpenAI co-founder Andrej Karpathy in a post on X, in which he described using AI-generated code for "throwaway weekend projects".

"I'm building a project or webapp but it's not really coding," he wrote. "I just see stuff, say stuff, run stuff and copy paste stuff, and it mostly works." Working with AI this way, "you fully give into the vibes, embrace exponentials, forget that the code even exists".

Code generation startups that develop software with generative AI are now "commanding sky-high valuations, as corporate boardrooms look to use AI to aid, and sometimes to replace, expensive human software engineers", said Reuters. Silicon Valley is seeing a "land grab situation" as the competing startups seek to "establish their AI coding tool as the industry standard". They're also rushing to beat the big companies, such as Google, Microsoft and Open AI from taking over the vibe-coding frontier.

For many tech companies, vibe coding is "a boon", said NPR. "They can now try lots of things quickly, rather than having to hand ideas off to software engineers to create prototypes, one by one." This means "Silicon Valley's next generation of unicorns" will not be restricted to those with the means to hire the "most talented coders", said Semafor.

It also means less work for coders, particularly those in entry-level positions who do more basic, repetitive tasks. Amazon CEO Andy Jassy said that last year the company had saved "the equivalent of 4,500 developer-years by using AI", said Reuters. In May, Microsoft announced that 20% to 30% of its code is now AI-generated; that same month, it laid off 6,000 workers.

'Messy' and 'full of bugs'

But, while "writing code based on vibes" can "feel like a superpower", said US computer science professor Chetan Jaiswal on The Conversation, AI-generated code still has serious limitations. Large-language models operate without context or adherence to "the processes behind developing production-grade software, which must balance trade-offs between product requirements, speed, scalability, sustainability and security". That doesn't matter for hobbyists but, for professional work, trained software engineers will still be needed to review and test the code before it goes live, especially if the software will be handling sensitive data.

Vibe coding may seem like magic on the surface, said the Financial Times, but, to a professional eye, it produces "messy code" that is "full of bugs and security weaknesses", and amateurs "do not have enough knowledge to spot all of these problems, let alone fix them".

It's a little like DIY: many of us will "experiment with projects at home and enjoy the process", and some of us "may even become really good" at it. "But, for the complicated jobs, many of us will discover a newfound respect for the professionals".