Vibe Coding: When Intent Becomes the Interface

THE SECOND SHIFT: FROM TYPING TO JUDGING

This leads us to the second major shift: the changing nature of “work” itself in a vibe coding environment. Stephenson described this change as a move into “reviewing…rather than…writing.”

He is not talking about lazy developers or cutting corners. He is describing leverage. In traditional development, senior engineers might spend 60% of their time typing scaffolding code and 40% thinking about architecture. Vibe coding flips that ratio. When an agent can quickly generate plausible code, UI scaffolding, and database schemas, the human role ascends the abstraction ladder.

The work becomes defining what “right” means. It becomes a task of judgment. The human must spot false assumptions, insist on security tests, demand documentation, and enforce safe defaults. The agent provides the draft; the human takes the responsibility. Fortunately, enterprises can leverage their existing code review and change advisory and quality assurance practices to complement vibe coding, even if the people who initiate reviews and engage in them change.

This is a more cognitively demanding task than many realize. It is often easier to write code from scratch than to debug someone else’s logic. In vibe coding, the human is constantly debugging an alien logic, one that is highly competent but prone to confident errors, and expressed in a language over which the human has little mastery.

This reflects a learning opportunity for organizations. Rather than viewing generated code as a roadblock, it can become an opportunity for learning. Most no longer understand the underlying principles upon which business tools are built. A SaaS application is not as transparent as a machine on a factory floor. By exposing the underlying building blocks, the creator can see how their idea is implemented, learn from the experience, and better communicate their intent in future revisions and refinements.

Nik Kale, principal engineer and product architect at Cisco Systems, put it this way: “Vibe coding is very productive when the problems you’re trying to solve are well-bounded, and you can clearly define the constraints (what are the inputs/outputs, error-handling conditions, security boundaries, etc.) of the problem. The biggest part of your productivity comes from using the model like a junior engineer that generates many options quickly while you hold the reins on the architectural decisions and correctness of the solution.”

THE THIRD SHIFT: CITIZEN DEVELOPERS

The migration story from Vendasta provides a perfect case study of the friction between “vibe” and “production.” Calista Cooper, director, AI business enablement at Vendasta, shared a familiar enterprise scene: a mountain of documentation scattered across various legacy systems and a corporate mandate to consolidate it all.

At first, Vendasta asked an R&D developer to build a migration tool to handle the heavy lifting out of Zendesk. The upfront investment was pretty low. Cooper noted that it “cost around $1,000” to get the tool built and running. While the tool moved content quickly, it didn’t reflect the team’s intent and therefore failed to return the expected output.

“It was actually a little bit of a hot mess because AI didn’t do the job perfectly. It hallucinated,” Cooper said. For instance, the migrated content “talked about product features we didn’t have.”

As Cooper’s team learned, any AI-driven migration without an automated verification loop creates an audit debt. The process would have benefited from content difference sampling or rule-based checks to avoid this problem. The lack of automated verification, however, was not without its upside. The resulting 6-week audit also revealed content left behind in another third-party repository. This allowed Cooper and her team to embark on a learning journey to engage in their own vibe coding work. They moved that content by employing Cursor and an IDE agent over about 3 days.

“It was the coolest feeling. The power you feel with this technology—[it’s] like you have a retention lever. Honestly, for people who are working here and getting a chance to do this—that feeling, I…boxed it in, and I [wanted] to feel like this all the time.” Cooper was feeling the “good vibrations” side of vibe coding.

Cooper and her team needed to build an entire system around their content. So rather than returning to R&D with a specification in hand, her team created a citizen developer program that empowered people to build using AI.

“It accidentally turned into this 3-month program. We learned about code literacy, AI literacy, provisioning, Google Cloud Build, IDEs, Markdown, machine learning, JSON, and, most importantly, how systems talk to each other. It completely changes the way you solve problems,” Cooper shared.

And at its heart, that is the “vibe” part of vibe coding. AI as collaborator, and, in the best of circumstances, humans “vibing” together with others to leverage the new tools. Barry Kunst, VP of marketing at Solix Technologies, calls this “builder ownership.”

Cooper sees it as “a different way of thinking. That’s something I didn’t expect. It’s completely changed the way that I work along with the other 20 people who are in this program. It brings us closer to [the product team’s] process. And importantly, it has brought the functions much closer together.”