Vibe Coding: When Intent Becomes the Interface
PRACTICAL ADVICE: PROMPTING AS REQUIREMENTS DISCOVERY
How do organizations get started with vibe coding? The answer lies in how they prompt.
Cooper shared a validation loop that changed how she learned to build. She didn’t start by trying to be a developer. She started by being stubborn about the outcome and letting the tool tell her what she didn’t know. Her technique is simple but profound: “At the end of every prompt, I’d say, ‘What more do you need from me?’” This one habit turns prompting into requirements discovery.
Vineel Arekapudi, principal data engineer at Wells Fargo, adds that prompting from written requirements is a good place to start: “Come up with a business requirements document that serves as a blueprint for the application. LLMs [large language models] frequently lose context, and having this document helps the model maintain context without a developer having to prompt it.”
In a traditional waterfall process, a business analyst might spend weeks interviewing stakeholders to gather requirements before a line of code is written. Vibe coding often skips this step, assuming the AI “gets it.” It usually doesn’t.
By asking, “What more do you need from me?” Cooper forces the agent to stop guessing and start asking. It forces the agent to surface missing dependencies: Do you have the API keys? What are the user permissions? What is the output format? What are the rate limits?
This technique forces the human to decide what belongs in scope. It transforms the interaction from a command (“Build this”) to a negotiation (“Here is the goal, what are the constraints?”). It turns the “vibe” into an active specification.
The biggest takeaway from my interviews was the need to have a vision for what you want to build, the skill to articulate that vision in a way that an LLM can understand, and the willingness to dip your toes into the developer world to transform a vision conjured by an AI into something more than a prototype.
Vibe coding won’t account for security frameworks or privacy policies unless the organization forces it to. AI demands visible guardrails. If citizen developers are contributing, the organization needs clear rules about what can and cannot go to production, what reviews are required, and how tests and documentation are handled. If the agent writes more, humans should spend more time validating, not less.
VIBE RE-ENGINEERING: THE FUTURE OF THE SME
As we look toward the future of this practice, it becomes clear that “vibe coding” is perhaps too narrow a term. It implies the end goal is always software. But in the enterprise, software is just a means to a process.
Anant Kale, CEO of AppZen, recently inspired a different way of looking at this evolution. During our conversation, I observed that we aren’t just vibe coding; we are entering an era of “vibe re-engineering.”
Vibe re-engineering happens when subject matter experts (SMEs)—the people who know the business deeply but the code rarely—learn how to co-create prompts that drive efficiency and innovation into processes.
In the traditional model, if an operations manager wanted to fix a broken workflow, they had to submit a ticket to IT to change the ERP system. In the vibe re-engineering model, the operations manager uses an agent to glue together the gaps in the process. The agent might “vibe code” a script that automates a daily reconciliation report.
Here is the “re-engineering” part: As the SME iterates with the AI, they are forced to examine the process itself. When the AI asks, “What logic should I use to approve this expense?” the SME might realize, “Wait, our approval logic is contradictory.”
The act of explaining the “vibe” to the AI exposes the inefficiencies of the human process. The SME isn’t just building a tool; they are re-architecting the workflow. They are using the rigor required by the AI to clean up the organization’s messiness. Vibe coding rewards people who can articulate outcomes, constraints, and edge cases.
The ROI of vibe coding lies in empowering the people who know the business best to act as architects, enabling them to enhance value by building solutions more rapidly.
Like much of AI, immediate productivity gains attract the most attention. Vibe coding contributes to the long tail of value driven by continuous improvement. Most organizations don’t instrument process change well enough to see compounding gains.
When intent becomes the interface, the barrier to improvement disappears. With vibe coding, we move from a world where we work around our systems to one where we co-create them continuously, conversationally, and collaboratively.