AI is changing how developers write software, and vibe coding has quickly become part of that shift. First emerging earlier this year, the method uses prompts to generate programs almost entirely through AI. It can speed up projects and open the door to coding without years of training, but it also raises concerns over reliability, security and oversight. Despite those challenges, its use among developers continues to grow, leaving technology leaders to weigh its promise against its risks.

“[With] vibe coding I’m putting in only prompts … and I’m letting my models write all of my code,” Sabrina Goldfarb, senior engineer at GitHub, said recently at Datadog’s DASH conference in New York. “I don’t have to be bogged down with technicalities and in writing syntax anymore. I can just focus on high level architecture and design.”

Vibe coding took hold in industry circles earlier this year after AI researcher Andrej Karpathy posted on X about a concept where developers rely almost entirely on AI to build software. Through it, they delegate most of the work to large language models, directing a process rather than programming, he wrote.

While vibe coding shows potential to transform how developers work, it has been somewhat controversial: some fear it may displace jobs, while others worry about the possible proliferation of  “buggy code” and a lack of experienced developers to build software.

The momentum behind it, however, keeps accelerating, as developers use it to cut down on repetitive tasks and prototype more quickly.

Prototyping At Scale

Diamond Bishop, director of engineering at Datadog, said vibe coding helps “write the code that I want … rather than a lot of the boilerplate that sits around it.”

Others at DASH lauded vibe coding’s ability to run multiple workflows at once, allowing developers to select the output that works best. It’s a tactic OpenAI’s Anoop Kotha compared to working with 25 engineers at the same time.

Beyond parallel workflows, teams are vibe coding to sketch out early ideas before committing to production.

“I think of vibe coding as more of a ‘let’s quickly put something out that we want to test out — prototype,” said Bishop. He noted that product managers are experimenting with tools like Codex or Cloud Code to assemble rough versions of products and features they can show to others. GitHub’s Sabrina Goldfarb said she’s been able to spin up full side projects in a weekend, most of which “just live on localhost,” not production servers. The goal is proof of concept.

Building Guardrails

AI tools might help build demos quickly, but high-stakes use cases require care and human oversight.

“If it’s … education or medical related, you need very rigorous testing systems to decide it’s production ready,” Luka Chkhetiani, staff researcher and research lead at AssemblyAI, said. He said teams should define clear boundaries early, including what the system should do, which risks are acceptable, and which failures are not. Without that kind of human judgment, they risk putting unstable or unsafe code into production.

Others say human review of vibe-coded output is needed to meet compliance and security standards.

“The moment you handle user data — even something as basic as an email — you step into legal and ethical territory,” Lindsey Witmer Collins, CEO and founder of WLCM App Studio and Scribbly Books, recently wrote in Fast Company. “Most vibe-coded apps aren’t designed with secure auth flows, proper data storage, or access controls.”

Refining Ideas In Real Time

One way to think about vibe coding is as a quick way to test ideas and adjust based on how they perform. Martin Amps, a member of the technical staff at Anthropic, described it at DASH as an iterative loop where feedback drives development.

The goal is to try to release a proof of concept early and use real-world reactions to shape what comes next.

“Generally, I’m fine shipping something with vibes,” he said, “and then I’ll continue iterating until there are complaints about regressions. As soon as somebody’s complaining, I have my first test case.”

The question now isn’t whether vibe coding will stick, but how soon teams can fine-tune early experiments into daily workflows.

Vibe coding is becoming part of daily developer practice, and the real challenge now is in guiding it responsibly. Leaders who encourage rapid prototyping while embedding the right guardrails for security, compliance, and oversight will be best positioned to turn early experiments into lasting value.