Why ERP is not just another platform you can rebuild with AI code

Vibe coding as a concept isn't new anymore. Enterprises are now exploring AI-assisted software creation and customization beyond the prototyping stage and confronting the operational realities that follow. 

There's a common misconception that organizations can easily rebuild core enterprise systems on their own because of how quickly AI can generate applications. For simple use cases like lightweight integrations or internal reporting tools, that may hold true. For systems of execution like Enterprise Resource Planning (ERP), the reality is very different.

A conversational shift must take place around AI in ERP, especially in regulated industries like manufacturing. Most organizations position AI as a feature inside ERP platforms, and that framing alone can be limiting and risk value. The real focus should be around AI as a production method for building ERP-like systems, extensions, customizations and even shadow ERP processes.

Put simply, AI is effective for acceleration, but anything that becomes a production dependency requires an actual operating model, not an AI prompt.

Where Vibe Coding ERP Falls Apart

AI-driven development removes friction, allowing teams to generate functionality quickly and iterate without being slowed down by architecture, standards or long-term considerations. It's valuable in the right context, advancing experimentation and lowering the barrier to building useful tools.

The problem is that manufacturing ERP is not a context where those tradeoffs hold.

 Manufacturing companies don't run on code alone. They must remain accurate under load, audit and change. ERP holds data accumulated over decades, encoded into processes that govern how a business actually runs. This data defines how inventory is valued, how production is scheduled, how costs are calculated and how traceability is maintained.

When organizations attempt to recreate that with AI-generated code, they are essentially attempting to replicate years of embedded logic, regulatory alignment and operational refinement, without the benefit of having lived through the edge cases that shaped it. 

The reality is, AI makes it easier to start building, but it doesn't make it easier to understand what you've built, govern how it evolves or ensure it behaves correctly under pressure. 

The Risks Leaders Aren't Pricing In

The looming mystery in this process is what happens after AI-generated software produces an output. The model works well when the requirement is simply, "does it work right now?" 

In manufacturing environments, success is measured far beyond the point of output: at month-end close, during an audit, when a supplier fails or when production needs to scale under constraint. Navigating successful implementation within manufacturing requires both hindsight from past operational realities and informed foresight about how systems must perform under pressure. This prototype-to-production gap is where most AI-built systems start to struggle, and a layered set of risks begin to compound:

• Code quality and security: One report found that 45% of AI-generated code contains security flaws. Researchers affiliated with Georgia Tech SSLab have also identified a growing number of vulnerabilities tied directly to AI-generated code. When that code is tied to inventory movements, production workflows or financial data, vulnerabilities directly affect business continuity and trust.
• Verification debt: Code generated quickly without clear specifications, structured testing or architectural discipline creates a backlog of validation work and maintenance. Over time, systems become harder to debug and more resistant to change.
• Unclear responsibility: When a custom, AI-generated system breaks during a critical process, there is no vendor roadmap, external ecosystem or clear internal ownership. The organization inherits full responsibility for maintaining and evolving something it may not fully understand.

When Governance Breaks Down, Operations Feel It

In manufacturing, these risks do not stay contained within IT. They show up on the shop floor, in production delays and compliance issues, leading to breakdowns in trust across the organization.

Mature engineering practices have always accounted for this. Secure development frameworks and specification-led approaches exist to ensure that what gets built can be tested, governed and sustained over time. 

Adding AI into the mix doesn't remove that requirement. If anything, it increases the need for discipline while making it easier to bypass. Without clear specifications, validation standards and ownership models, there isn't a mechanism to ensure the right governance is applied. 

Decisions are made in-the-moment, with speed often taking priority over structure, leaving security as an afterthought. Over time, that erodes consistency, introduces risk and pushes security and reliability into the background until they become visible through failure.

Don't Abandon AI - Apply It With Discipline

The efficiencies AI-generated code can provide are real, and ignoring it would be a mistake. The mindset must shift from generation to specification. AI should operate within defined parameters, clear intent and established architectural guardrails. Instead of asking "can we build this quickly?" the better question is "can we define this clearly enough to build it safely and scale it reliably?"

Focus AI investment on where it delivers the most value with the least risk to core integrity. Examples are already emerging across manufacturing environments: automating document processing, enhancing shop floor visibility, guiding user workflows and surfacing insights from operational data. These are areas where AI can accelerate outcomes without compromising the system of execution underneath.

At the same time, preserve what already works. Proven ERP platforms bring embedded logic, compliance alignment and a support structure that evolves with the business. AI can amplify that foundation.

There's no reason to disregard vibe coding entirely. Proceed with caution and apply AI in a defined direction within a framework that maintains control, consistency and trust. AI can accelerate the enterprise, but only if the foundation it runs on is one that can be trusted.