A developer proudly shows me his latest creation. In three weeks, he built an AI agent that aggregates reports from four internal systems. The demo works flawlessly. Management is impressed. Three months later, the same agent is in production and a ticket escalates: the numbers in the weekly board meeting no longer add up. No one can reproduce where the agent gets its data. The developer explains honestly: “I had that piece generated entirely by Copilot. It worked, so I left it that way.”
A story like this is rarely about a bad engineer. It is about an industry that is changing at breakneck speed, and about how to respond to that as an organization.
This is the fifth article in our series on data and AI. In the previous parts, we looked at AI wild growth in organizations, Power Platform as a bridge to controlled AI integration, the data foundation with Microsoft Fabric and grounded AI as an architectural approach. Each time, the focus was on the technology. This time we look at the people who work with it. Because without the right mindset, every technical investment remains an experiment that one day gets derailed.
The vibe-coded agent from our first article is not just a footnote. It is symptomatic of a fundamental shift in how software is created today, and how it must be maintained tomorrow.
How vibe coding came about
Earlier this year, Andrej Karpathy, former head of AI at Tesla and one of the most influential voices in deep learning, introduced a term that divided the software world into two camps: vibe coding.
His description was as simple as it was provocative. You surrender to AI, you describe what you want to build, you accept what it generates without fully understanding it, and you keep pushing until it works. “I’m just vibing,” he wrote. The code belongs to AI. You drive the intent. The reactions were predictable: enthusiasm on one side, horror on the other. Both reactions miss the actual question this way of working asks of anyone who builds software, deploys AI or makes technology decisions for an organization.
The right question goes deeper than discussing whether to accept code you don’t fully understand. The right question is what is fundamentally changing in the engineer’s role and how you build on that as an organization. Vibe coding is here. The question is how you deal with it.
Vibe coding is a technique. Vibe engineering is a mindset. It determines the difference between more productivity and more risk.
What vibe coding is and what it is not
Vibe coding is the practice where a developer deploys AI tools such as GitHub Copilot, Cursor, Claude or ChatGPT to generate code based on intent rather than implementation. You describe the desired behavior in natural language. AI writes the code. You review, test, adjust and repeat.
In its most radical form, as Karpathy described, you accept the output without delving deeply into the logic. You rely on the system rather than your own understanding of it. This works surprisingly well for prototypes and for situations where speed is more important than robustness.
So that’s where the criticism is justified. Code you don’t understand, you can’t debug. Code you don’t understand, you can’t secure. Code you don’t understand can’t scale to a production environment that serves thousands of users and has compliance requirements.
Vibe coding is a powerful tool. A tool without skill is dangerous.
Vibe coding versus vibe engineering: the real distinction
The difference is fundamental. Vibe coding is a technique that allows individual developers to produce faster. Vibe engineering is the discipline that ensures that that speed delivers value rather than technical debt.
Where vibe coding focuses on the output of an individual task, vibe engineering looks at the system as a whole. Where vibe coding accepts that code works without understanding, vibe engineering requires that architectural decisions be reasoned. Where vibe coding treats the AI as a resource, vibe engineering treats the AI as a colleague sitting at the table with strengths and weaknesses.
Vibe engineering builds on vibe coding. It is its mature version. It embraces the productivity gains offered by AI-assisted development and places those gains in a framework of craftsmanship, architectural thinking and organizational accountability.
The vibe engineer uses AI the way an experienced surgeon uses a scalpel: as an extension of expertise.
The five pillars of vibe engineering
What makes someone a vibe engineer instead of a vibe coder? It involves five shifts in thinking, method and responsibility.
- Intention over implementation
The engineer defines what the system should do and why. AI determines how. The engineer remains owner of the architectural choices and monitors that those frameworks are reflected in the generated code. - Critical reviewing as a core skill
Reading AI output is a skill. Vibe engineers know what they’re looking for: edge cases, security vulnerabilities, performance pitfalls and unwanted dependencies. Plausible code is different from correct code, and only a trained eye sees the difference. - Prompt engineering as an architectural language
A good prompt is a good specification. Vibe engineers learn to communicate accurately with AI, which also improves their communication with humans. Clear intent produces clear output. Vague briefings produce vague code. - Governance by design
Security, compliance and testability are included from the first prompt. They are part of the prompt itself, the review and the acceptance criteria. Retrofitting always costs more than pre-building.
- Continuous learning from AI
AI-generated code often reveals patterns and solutions that the engineer did not already know. Vibe engineers actively use that to deepen their own knowledge. The collaboration goes both ways: the human directs; the human also learns.
→ The result: engineers delivering faster and writing better code, thanks to thoughtful collaboration with AI.
Why this is important for your organization
This story is not just relevant to developers. It touches everyone who makes decisions today about how software is built in their organization.
An organization that embraces vibe coding without cultivating vibe engineering sees impressive productivity gains in the first three months. After that, the problems begin. Code that no one understands. Security vulnerabilities that no review has noticed. Systems that are not maintainable. It’s the same sprawl we described in the first article in this series, only one layer deeper, in the codebase itself.
Vibe engineering avoids that scenario. It makes AI’s productivity gains sustainable by having engineers use AI as an extension rather than an outsourcing partner. An assistant who codes with them instead of in their place.
For IT leaders: redefine what you expect from engineers
Tomorrow’s developer has less value as a typist and more value as an architectural thinker, critical reviewer and AI collaborator. Adjust your hiring, your training and your assessment criteria accordingly. Today.
For C-level: understand the productivity paradox
AI increases output rates, but without vibe engineering, it also increases your risks. Organizations that invest in the right practices around AI-supported development now will be the leaders in two years. Those that don’t will pay an expensive bill later in the form of technical debt, security incidents and rework.
For developers: this is your moment
The engineers who learn how to best use AI today as reviewers, architects and prompt engineers will become the most valuable professionals of the next decade. The bar is higher than ever. Those who grow with it will find opportunities that weren’t there before.
The question is no longer whether your engineers will use AI. That question has already been answered. The real question is how they do it: in a way that delivers value or increases risk.
The broader implication: a higher level of abstraction
Vibe engineering is symptomatic of a broader shift we are seeing across the industry: the abstraction layer is rising. Just as we once stopped writing assembler and moved to higher programming languages, we now stop writing every line ourselves and start operating at a higher level of abstraction.
Craftsmanship does not disappear here. It evolves. The best engineers of the next generation will be judged on the quality of the systems they build, regardless of what tools they use to do so. The number of lines of code they personally write weighs less and less.
That layer of abstraction does require something new. A deeper understanding of why systems work, on top of knowing how they work. When you delegate implementation to AI, you can only make architectural choices if you understand the underlying principles.
That’s what we at Xylos mean by vibe engineering. For us, it’s about a way of working that takes full advantage of the potential of AI, with the human expertise that remains irreplaceable as an anchor. The hype with it is secondary.
Xylos helps your engineers evolve to the vibe engineering mindset
The transition to AI-assisted building is more than a tooling issue. It touches culture and skills. At Xylos, we guide organizations through that transition: we help select the right AI development tools like GitHub Copilot and Azure AI, set up review and governance processes, and retrain engineers in the new core skills of the trade. Want to know where your organization stands today? Contact us for a free consultation.
About the author
Peter Verrykt is Business Unit Lead Data & AI at Xylos and guides organizations in turning data into concrete business value. He helps companies look beyond technical implementations and deploy data and AI as a foundation for better decisions, greater agility and sustainable growth.
