Vibe Coding: Move Fast, Break… Everything?

Vibe coding is the practice of describing what you want in plain English and letting AI generate the code for you.

It is gathering serious momentum. No syntax. No boilerplate. No wrestling with APIs. Just intent, and an app.

On the surface, it feels like magic. And to be fair, in many respects it is. It democratises software creation in the same way low-code did a few years ago, arguably even more so. People who have never written a single line of code can now build something functional. That is not a trivial development.

But here is the part that tends to get skipped over in the excitement. We have seen this movie before.

Vide Coding is Impressive

There is real merit here. The speed at which vibe coding allows you to move from idea to working prototype is impressive. As mentioned, vibe coding lowers the barrier to entry massively, shifts the focus to outcome, and for throwaway apps, early-stage experiments and rapid proof-of-concept work, it represents a meaningful step change. The traction it is gaining is not hype for hype's sake and when used in the right context, it is a powerful tool.

The problem begins when people assume universal applicability.

The Misconception

The fundamental misunderstanding about vibe coding, and low-code before it, is the assumption that software engineering is primarily about writing code. It isn't. Writing code is one relatively small part of a much larger discipline. Real-world enterprise systems are defined by architecture, data models, integration patterns, performance characteristics, resilience under load and governance frameworks. That is where the genuine complexity lives, and that is precisely where AI-generated code begins to wobble.

Take integration as a starting point. Vibe coding is excellent at generating something in isolation. It is considerably less impressive when that something needs to plug into a messy enterprise landscape built on legacy systems,fragmented data, strict API contracts and layered security and identity infrastructure.

AI-generated code tends to struggle when it must align to existing architectural constraints, often resulting in duplicated logic, disconnected data and fragile point-to-point connections. In other words, it is shadow IT, just faster.

No Clear Source of Truth

Data modelling rarely makes it into the vibe coding conversation, and that is a serious oversight. Enterprise systems live and die by their underlying data model, its relational integrity, its ability to enforce rules, its capacity to evolve gracefully over time. AI-generated applications tend to optimise for the immediate requirement, producing structures that work well enough right now but become difficult to unpick once multiple users depend on them, reporting becomes critical or data quality starts to matter.

What begins as a neat, self-contained solution quietly becomes a spaghetti of loosely defined structures with no clear source of truth.

Scalability compounds the problem. AI-generated code works brilliantly, right up until it doesn't. Many generated solutions assume small data volumes, rely on inefficient queries or processing loops and lack any meaningful optimisation for real-world load. The demo works beautifully. The production environment at scale is a different conversation entirely, and at that point the responsibility for performance tuning sits firmly with the humans who deployed it, not the AI that wrote it.

Risk Profiles and the Question of Resilience

In enterprise environments, downtime is not an inconvenience, it is a financial and operational risk with very real consequences. Platforms like Microsoft's Power Platform are built with this in mind, offering multi-region replication, automated backups and near-zero downtime failover as standard.

With vibe-coded solutions, the answers to basic resilience questions such as:

·      Where is this hosted?

·      What is the failover model?

·      How is data recovered?

are frequently unclear, inconsistent or dependent on manual configuration that may or may not have happened.

That is a fundamentally differentrisk profile and one that tends to surface at the worst possible moment.

Technical Debt at AI Speed

Traditional technical debt accumulates gradually, giving organisations at least some opportunity to recognise and address it (albeit it still becomes a challenge for most). AI-driven development scales debt at precisely the same speed it scales delivery. Research consistently shows that AI-generated code is often functional but frequently lacks the architectural judgement that distinguishes a working answer from the right answer for a specific system. The result is duplicated patterns, inconsistent design and code bases that become someone else's problem to maintain, usually sooner than anyone anticipated or planned for.

What About Low-Code?

Low-code is where the conversation becomes genuinely interesting for those of us working in the Microsoft Power Platform space. Properly implemented low-code occupies a valuable middle ground: faster than traditional software engineering, more controlled than vibe coding and built with enterprise guardrails from the ground up.

Structured data through Dataverse, managed integrations, robust governance and security models, and built-in scalability and resilience, are not afterthoughts bolted on later. They are foundational. And critically, when the situation demands it, you can still drop into code. Low-code enables speed, without abandoning structure entirely.

The Key Question

Vibe coding is exciting. It is fun. It absolutely has a place in the modern development toolkit. But it is not a replacement for software engineering, and it is not a free pass on complexity. If anything, it accelerates towards complexity rather than away from it, creating a real risk of breaking things.

The question worth asking is notwhether you can build something with vibe codingalmost certainly can. The more important question is what happens when the thing you’re building iscritical to your operation – i.e. it is something that needs to integrate withcritical systems, scale under real demand, meet security requirements and stayrunning when the organisation depends on it.In other words - when it is something thatcan’t break.

By all means, embrace the vibe. Vibe coding is a real step forward in making technology more accessible for all and there is something liberating about the speed and creativity it enables. Just keep one eye on what sits underneath it.

The code might be invisible. The consequences are not.

If you’re figuring out what vibe coding means for your organisation - where it helps or hurts - we’d love to talk. Get in touch with us today to arrange an introductory call.