Earlier this spring, Alex Balazs hosted an internal hackathon at business software provider Intuit, where the chief technology officer challenged his global team of engineers to embrace artificial intelligence in their work as much as they could.
To give them the tools to do so, one week before the biannual meeting, Balazs approved the usage of “every AI coding tool that we’re aware of,” including Qodo, Windsurf, and Cursor. He quickly saw that thousands of Intuit’s engineers were using these tools, up from hundreds previously.
“Our metrics are showing us that engineers love it,” says Balazs. Intuit is tracking how frequently the AI coding tools are being used, the steadiness of that usage over time, and productivity. Early indications are that efficiency gains can be as high as 40%.
Along the path to going all-in on AI coding assistant tools, Balazs finds himself enticed by a new concept called “vibe coding.” It is a term that was coined earlier this year by OpenAI cofounder Andrej Karpathy where users describe a project’s goals in simple, natural language to an AI-powered coding platform like Windsurf, which then writes the code.
Balazs recently experimented with vibe coding when using Windsurf to create a tool that would allow him to import data from a competitor product directly into Intuit’s QuickBooks software. “I found myself making significantly more progress with my rusty coding skills than probably I would have otherwise,” says Balazs.
The extent to which the “vibe coding” hype ultimately turns out be a real force that upends workflows across the business world remains to be seen. But the industry’s largest AI hyperscalers are paying very close attention. OpenAI is reportedly in talks to acquire vibe-coding software startup Windsurf for $3 billion. Apple and Anthropic reportedly joined forces to develop their own AI-powered vibe-coding platform. And more and more code is already being written by AI: at Microsoft, as much as 30% today.
Better reasoning models, which can understand the nuances of natural language and translate that into code, are a key factor propelling the trend. “This past winter is where the vibe coding wave really started to take off,” says Alex Albert, head of developer relations at AI startup Anthropic. Newer reasoning models, like Claude 3.7 Sonnet, are enabling developers to take a more hands-off, managerial approach.
Walter Sun, global head of AI at German business software firm SAP, says the conversation swirling around vibe coding reminds him of the earliest days of his career. His older peers would tell him he wasn’t a true developer because Sun didn’t do punch card programming, the use of physical cards to instruct early computers. “I believe that vibe coding is just a further evolution of productivity for development,” says Sun.
He says the principles of vibe coding, along with the broader use of AI-enabled coding tools, is an encouraging trend for developers who are creating generative AI tools to eliminate mundane tasks for the broader workforce. But they should also be thinking of ways to use AI to make their work more efficient. “I think that it’s actually great that developers are drinking their own champagne,” adds Sun.
In March, SAP rolled out an AI-enabled tool for the company’s developers, intended to accelerate workflows by performing code competition and offer code explanation to older forms of code that newer developers may not be familiar with. This tool embraces the principles of vibe coding, Sun says, and internal analysis shows that code is being produced 20% faster than the rate before this tool was rolled out.
Albert says Anthropic is currently having sales conversations with enterprise customers to explain what vibe coding is. Anthropic has done this type of education around AI agents, explaining what these systems are, when and when not to use them, and the frameworks necessary to implement them. “When a company says, ‘I want to lean into vibe coding,’ it is more about how we can give our software engineers more power and more control over these AIs and allow them to build way more, way faster,” says Albert.
Todd Olson, cofounder and CEO of Pendo, says vibe coding has helped inspire him to take up coding again. “It is harder for me to lead the business if I don’t have first-hand experience with what these solutions do,” says Olson. “It allows me to set a strategy much, much easier.”
Roughly a third of the software company’s 900 employees are engineers and vibe coding is permeating across the broader workforce, mostly helpful in the earliest stages of software development, including prototyping. With a more iterative process, Olson expects that Pendo and others can rapidly accelerate the software development cycle. “Historically, it has been hard and slow to get software out,” says Olson. “We’re all just going to test more.”
Technologists say that vibe coding won’t replace all of the work of the engineer. As an example, all AI-generated code that Pendo may use during the software development lifecycle doesn’t go right into production. There is still an employee review process to ensure that the code that’s AI-generated has no security loopholes and adheres to Pendo’s coding standards.
That said, AI coding tools are already changing workflows. The primary role of a software engineer is tilting more toward editing code, rather than writing it.
“Technology has been happening for literally thousands of years,” says Balazs. And each time it occurs, he argues, workers must make a choice to either adapt or not. “All the best engineers will be using these tools,” he added. “So if you want to be one of the best engineers, use these tools.”
John Kell
Send thoughts or suggestions to CIO Intelligence here.
This story was originally featured on Fortune.com