Vibe coding, or using natural language prompts to generate software, allows developers and non-specialists to create applications with AI assistance rather than writing every line of code manually. Clearly, thisaligns very closely with the emergence of GenAI, to the extent that, according to GitHub, AI coding assistants generate around 46% of the code written on its platform.
Elsewhere, the trend is even more pronounced, with 2025 research from Stack Overflow revealing that 84% of respondents either use or plan to use AI tools in their development process, up from 76% the previous year. The appeal is obvious. At a headline level, AI reduces the friction associated with software development and lowers the barriers to creating useful tools. What’s not to like?
However, this has led to a growing belief that because AI can generate applications so quickly, organisations can rebuild or replace existing enterprise systems themselves. Granted, this may hold true for relatively simple use cases such as internal dashboards and lightweight integrations, but the picture changes considerably when the software involved sits at the heart of business operations.
The challenge of production systems
Among the most infamous recent examples are the problems encountered by PocketOS, a SaaS startup that had successfully established itself with just three founders and no software engineers. They are, in their own words, “all in” on AI. It was a strategy that appeared to be working well until one of their AI agents went rogue and deleted the company’s entire production database and backups.
As PocketOS reported at the time, this existential process took just nine seconds. Thankfully, the business was eventually able to recover, but it’s already become a case study for what can happen when AI-generated development outpaces the operational disciplines required to sustain it.
While that’s among the most extreme examples, every market niche is potentially vulnerable. Take manufacturing environments, for instance, where ERP systems sit at the very heart of how businesses operate. They provide the operational backbone that coordinates and manages core processes, from procurement and inventory to production, finance, and traceability.
They are also, however, much more than collections of code. They represent systems of execution, making the implications of AI-generated development very different from the fundamentals that drive vibe coding.
ERP platforms contain embedded logic that determines how key processes operate, with much of this complexity resulting from accumulated operational experience. In many businesses, edge cases, exceptions, and regulatory requirements have shaped these systems over many years or even decades.
Success in manufacturing is measured over time and under pressure. Systems are expected to continue performing during audits, month-end close, supplier disruptions, and periods of growth; the list goes on. This creates a gap between building software quickly and operating it reliably. It is often only when systems encounter real-world complexity that weaknesses become visible.
In this context, generating code with AI is not the same thing as replicating the operational knowledge and business rules that support established ERP environments. Yes, AI can make it easier to start building applications, but it does not automatically make them easier to understand or maintain, particularly as requirements evolve.
Applying AI with discipline
The question organisations should ask is not whether AI-generated software works today, but whether it will continue to perform under changing business conditions. In manufacturing, success is often judged at moments of stress, such as month-end close, audits, supplier failures and production bottlenecks.
The development challenge is that rapid code generation can create verification debt. Systems that lack clear specifications and structured testing become progressively harder to debug, modify and maintain. Speed can also obscure accountability. When problems occur in custom AI-generated environments, organisations may find themselves wholly responsible for systems they do not fully understand.
These issues rarely remain confined to IT. Any weaknesses may eventually manifest operationally through delays and a loss of confidence in the underlying systems. The challenge, therefore, is not whether AI can generate software quickly. It is whether organisations can ensure that it remains reliable and governable long after the initial build phase has ended.
Vibe coding, but with a different mindset
Make no mistake, mature software engineering practices exist for a reason. Specification-led development, testing frameworks, governance processes, and a range of other disciplines were created to ensure systems remain sustainable over time. Introducing AI into development does not remove the need for those disciplines. If anything, the speed AI provides makes them more important.
The mindset should shift from asking “Can we build this quickly?” to “Can we define this clearly enough to build it safely and scale it reliably?” AI should operate within established architectural guardrails and ownership models rather than replace them. Organisations should focus AI investment where it delivers value without compromising the integrity of core systems.
But let’s also be clear, there is no reason to reject vibe coding. The challenge is ensuring that AI-driven speed is matched by the control, consistency, and trust required to support production environments.
In manufacturing, proven ERP platforms already provide embedded business logic and support structures that evolve alongside the business. AI can and should amplify these foundations rather than attempt to recreate them.
