There is a fascinating, slightly intoxicating idea circulating through executive suites right now: Vibe Coding or AI-assisted development. Armed with a sophisticated Large Language Model (LLM), a motivated developer can now spin-up a preliminary framework and an impressive-looking Industrial IoT dashboard over a single weekend. It feels fast, it feels cheap, and it makes the traditional “buy-or-build” calculation look instantly obsolete. But behind this illusion lies a dangerous strategic trap: the belief that because AI can generate code in minutes, your company should build its own underlying IoT infrastructure from scratch.
The real threat to OEMs over the next years isn’t that a competitor will successfully vibe-code an enterprise IoT platform offering. The threat is that your team will believe they can, and will waste years trying to convert a prototype into production-ready industrial infrastructure. In a modern industrial market governed by NIS2 and the EU Cyber Resilience Act, an un-authored, custom-built infrastructure is no longer just technical debt; it becomes a compliance and operational risk that executives must own.
Here is the reality check on vibe-coding your underlying IoT infrastructure, and how forward-thinking OEMs are shifting the narrative to "buy and build faster.
The Excel Macro Era, All Over Again
We have watched this movie play out before. On a recent episode of The IT/OT Insider podcast, the discussion highlighted that the current wave of generative AI experimentation looks a lot like history repeating itself.
“Ten years ago, everyone was stitching together critical business processes using custom VB macros in Excel and fragile Access databases. Today’s version is the enthusiastic developer spinning up open-source modules or tying together a dozen raw hyperscaler components natively in a cloud sandbox. It feels like ultimate freedom, but it completely ignores the looming hidden costs.”
- David Ariens @ IT/OT insider
Back then, those enterprising employees who built clever shortcuts created a massive long-term headache for operations. A single undocumented macro would run a critical workflow flawlessly, until that employee left the company. Today’s vibe-coded version risks the exact same flavor of failure. Your engineers might love spending time architecting that underlying data plumbing, but your customers don’t buy your machines because of your custom data routing implementation.
Base infrastructure is just the entry ticket to the market. The real battle to win a customer happens at the level of unique differentiation, where your team focuses its talent on building the proprietary, high-value features that set your connected machine offering apart from the crowd. Your customers buy outcomes: uptime, throughput, and cost efficiency.
The Reality Check
A prototype that visualizes telemetry data on a screen might look like a functional platform. But in the world of industrial machinery, a polished user interface is just the tip of the iceberg.
Surely, AI can accelerate isolated coding tasks allowing you to spin up a dashboard that visualizes temperature data in an hour. But what if that same system needs to manage 50,000 devices across 20+ countries, orchestrate OTA updates without bricking a fleet, and satisfy auditors under NIS2? The real risk of vibe coding is mistaking a convincing user interface for a resilient system that can actually serve as a system of record, as a trusted foundation.
Beneath the surface sits the mission-critical infrastructure responsible for secure device onboarding, managing tens of thousands of distributed assets, and orchestrating over-the-air (OTA) firmware updates. When you use an AI framework to spin up this infrastructure layer, it may successfully hand over a mountain of automated code. But then it wishes you good luck and disappears.
“A LLM will not sign your SLA with a customer. It will not take responsibility when things go sideways with operating the platform, e.g. regarding cybersecurity, 24/7 platform availability, and scalability.”
- Dr. Jürgen Krämer, CPO Cumulocity
We are operating in a post-NIS2 and EU Cyber Resilience Act (CRA) world, where connected machine fleets are classified as critical infrastructure. Maintaining a custom-built, unauthored codebase against strict contractual uptime commitments and heavy global cyber mandates isn’t just difficult—it’s an immense corporate liability.
Consider the operational reality:
-
The OTA Failure: A rushed firmware rollout, generated by a patched AI model, unexpectedly bricks 15% of your installed machinery across three different countries.
-
The Compliance Breach: A subtle, unreviewed vulnerability in your custom device onboarding flow triggers a massive data leak and a corresponding regulatory compliance breach.
When a crisis like that hits, an LLM will not hop on a support call, nor will it handle the legal fallout under strict NIS2 mandates. The responsibility lands squarely on the executive suite, not the AI tool. Running mission-critical industrial infrastructure requires a level of robustness, compliance, and accountability that a prompted script simply cannot guarantee.
Shift the Mindset: The Launchpad for Your Applications
A vital architectural distinction needs to be made: Vibe-coding or AI-assisted development is not a substitute for a proven platform at the foundation. It is however genuinely useful at the application layer—building custom dashboards, tuning models on your own data, and accelerating vertical use case development. It is an incredible superpower when applied there.
“To win in the modern era, OEMs must realize the answer isn’t buy OR build; it’s about buying a robust foundation so you can build your domain-specific applications FASTER and CHEAPER, with the help of AI.”
– Dr. Jürgen Krämer, CPO Cumulocity
This is exactly why we have made Cumulocity completely accessible to the broader AI ecosystem. By exposing comprehensive open APIs and natively supporting the Model Context Protocol (MCP), we provide the deep semantic context that any external AI agent or LLM framework requires to query true operational realities. It turns the underlying infrastructure into an open launchpad where developers can plug in their preferred AI tools and build apps in record time.
By buying this pre-built operational floor, your development teams can point their AI tools at a highly structured, open foundation and focus 100% of their energy where it actually belongs: on building the vertical applications, predictive models, and high-margin digital services that actually move the needle for your business, unlocking a completely different trajectory for your company:
-
Accelerated Product Launches: Instead of spending two years writing custom code just to safely connect your fleet, your team can use generative AI to roll out your initial connected product offering in weeks. Your time-to-market for smart assets shrinks from years to months.
-
Higher-Margin Services: Your machines cease to be commoditized hardware. With your data structured and open, your engineers can dedicate their focus to building unique, differentiating additions—such as premium digital subscriptions, outcome-based analytics, and predictive maintenance models that create predictable, recurring revenue streams.
-
Happier, High-Impact Engineering Teams: Your best R&D talent is no longer trapped in the exhausting grind of “maintenance hell” or fighting compliance paperwork. They get to focus entirely on the application layer and creating the innovative intellectual property they actually care about, driving the visible competitive advantages that help your business win market share.
This is the true promise of the AIoT era. You aren’t using AI to frantically rebuild the floor; you are using it to rapidly build a premium ceiling on top of a rock-solid foundation.
The Strategic Choice
The choice for OEM executives isn’t between embracing AI development or rejecting it; it is about where you direct that superpower.
Using AI to rebuild foundational infrastructure from scratch is a dangerous, multi-year distraction. It traps your team in maintenance hell, TCO spiralling out of control and leaves your company exposed to significant operational liabilities and compliance risks under NIS2 and CRA mandates.
A “buy and build faster” strategy takes a different path. You inherit global scale, built-in compliance, and an open, AI-ready foundation - both in terms of data and application building - from day one, so your teams can focus on creating value, not maintaining plumbing.
Don’t waste the next three years trying to replicate a mission-critical platform your competitors bought on day one. Use that time to differentiate.
Secure your fleet, unlock your data, and use AI to build what actually makes your business unique.
Buy the foundation. Vibe-code your innovation on top. Build faster.
We, at Cumulocity, operate and build on this exact same open architecture. We don’t just advocate for AI at the application layer, we even leverage AI where it actually removes friction. We looked into the operational bottlenecks that slow teams down, investigated AI-first solutions and embedded AI into those moments to accelerate time-to-value, for example:
-
AI-Powered Data Preparation: Instead of forcing engineers to manually write complex mapping scripts for every new asset type, this tool uses AI to easily translate and automatically map chaotic, raw machine data into the standard Cumulocity format. It handles the harmonization out-of-the-box so you can connect and parse data instantly. (For a deeper dive into how this works, see our companion article on automated industrial data prep.)
-
Natural Language Dashboards: Your developers no longer need to spend days configuring layouts and individual UI widgets. Because our platform understands natural language queries, users can simply type a prompt, and the embedded AI assistant dynamically generates the exact tailored dashboard they need. (Check out our blog on generative UI and dashboard generation to see it in action.)
Build What Differentiates You
AI is changing how software gets built. The winners won’t be the companies that generate the most code; they’ll be the companies that direct that capability toward the right problems.
Build the applications that differentiate your business. Leave the operational foundation to a platform designed to run at industrial scale.