Vibe coding is making software development easier to approach, even for people who are not traditional programmers. By using natural-language prompts, developers can ask artificial intelligence (AI) to generate code, test ideas, and quickly turn concepts into working prototypes.
But for enterprises, getting an application from an idea to a production-ready system requires more than generating code.
“Vibe coding will continue to grow, and we see that as an amazing thing because it will help, like if I have an idea, how do I bring that closer to where it is?” said Joel Garcia, technology lead, Strategic Initiatives, ASEAN, AWS, during a media workshop.
“It’s low barrier, it’s easy to use, it’s fast and again, they can validate and iterate on some of their ideas,” he said.
That accessibility could encourage more people to experiment with coding, even without a formal programming background. The challenge starts when an organization wants to use that code in a real business environment.
“As they kind of continue going forward and as their requirements grow, they realize that there’s more requirements similar to what we had talked about,” Garcia said. “They need to look at specification, they need to look at documentation, they need to look at testing requirements, and so on and so forth, that are required sometimes by their industry, sometimes obviously by their organization.”
As companies move from experimenting with AI-generated code to deploying it, there will always be the issue of security.
“There are specific guardrails and steering files that you give the AI to ensure that the output that you are trying to get to in terms of code, in terms of the application, meets the corporate standards, meets security standards, meets the enterprise standards,” Garcia said.
He also noted that security measures such as penetration testing is part of moving an application toward production. It means that writing code is only one part of software development. Enterprises also have to consider security, testing, documentation, and how an application will work at scale.
That is where agentic coding enters the picture.
Garcia said organizations are moving beyond simply generating code and asking how those applications can be deployed “faster, but in a scalable and secure fashion.”
AWS sees Kiro, which was previewed at AWS re:Invent 2025, as a way to bring more structure to AI coding. Garcia described it as “our agentic AI development tool.”
Kiro uses AWS spec-driven development. Instead of relying only on a broad prompt, developers can define clearer business requirements, system design, and the tasks needed to reach an intended outcome.
“Your first spec won’t be your final spec,” Garcia said, explaining that developers can continue to refine the requirements before the AI implements them.
The approach is also designed to keep people involved. Garcia emphasized that “even with agentic coding, there is still human in the loop,” with people reviewing, steering, and iterating on specifications.
That distinction matters as AI coding tools become more capable. This does not mean removing developers from the process, but to enable AI handle more of the execution while people focus on decisions that require context and judgment.
For beginners, that could make coding more approachable. For enterprises, however, the bigger challenge is making sure AI-generated software is secure, reliable, documented, and ready for production.