This article originally appeared in ICT/Magazine.

Software development is fundamentally changing because of AI. Twelve years ago, developers still wrote every line of code manually. Today, organisations can build complete applications, dashboards, or workflows with just a few prompts. “Vibe coding” sounds appealing: faster development, lower costs, and less dependence on major software vendors. But are these applications secure? Do they integrate with existing systems? And can they remain manageable once they become part of critical business processes? This fundamental shift in software development forces organisations to make fundamental choices: should we build software ourselves? And if so, how can we do it responsibly?

SaaS vs custom development

What we are seeing now with AI is a renewed shift in the traditional buy-versus-build discussion. For years, organisations opted for large software suites under the assumption that standardisation would be more efficient. In practice, however, not all standard functionality proves relevant, and in many cases it is barely used. As a result, organisations still require extensive customisation to make processes work properly.

Take SAP as an example. Many implementations started with the promise of “100% straight-through processing”. In reality, those numbers are often much lower, leading to custom-built solutions inside standard software packages. This is not only expensive, but also creates manual workarounds, additional Excel sheets, and operational inefficiencies.

AI is changing that balance. Custom software is no longer inherently expensive, complex, or time-consuming. Organisations can now build solutions that align far more precisely with their processes, sometimes as an extension of the existing application landscape, and sometimes as a replacement for existing applications. This gives organisations more flexibility and allows them to adapt more quickly to changing business needs.

Perhaps the most interesting consequence of AI-driven software development is what it means for legacy systems. For years, many organisations depended on outdated applications nobody dared to touch anymore, the classic black screens with green text. Systems containing critical business logic, often maintained by a handful of specialists nearing retirement. In the past, modernising these systems was often too risky or simply too expensive. In 2026, AI models can analyse existing business logic, structure documentation, and even generate new user stories based on legacy applications. Suddenly, knowledge trapped inside outdated systems becomes accessible again and can be translated much faster into modern applications.

The next step: Agentic Application Development

Vibe coding is a new technological wave that strongly reminds me of earlier waves such as low-code development. At the time, people also questioned whether that technology was suitable for enterprise applications. We now know those doubts were unfounded. Every promising new technology only succeeds when applied intelligently: with the right frameworks and the right expertise.

What remains unchanged are the requirements organisations place on their application landscape. Applications still need to be secure, scalable, and manageable. They must integrate with existing systems and, above all, deliver measurable business value.

What I increasingly see in practice is teams independently building tools and applications without clear frameworks or sufficient expertise. Initiatives start within departments without alignment with IT. They may work for a small group of users, but they rarely integrate properly with the right source systems, not to mention the quality and security risks involved. Some companies even refer to these as OEIs: unwanted entrepreneurial initiatives.

This quickly leads to a fragmented IT landscape. Applications function in isolation rather than together. Data becomes scattered, governance is missing, and maintenance grows increasingly complex. In short: the foundation starts to crack.

To move beyond pilots and experiments with vibe coding, organisations need more than a fast prompt or a working prototype. They need an approach in which the business challenge, architecture, integration, security, and maintenance come together from the very beginning. At Ciphix, this is what we call Agentic Application Development.

Agentic Application Development means applying AI within the frameworks required to develop and manage enterprise-grade software. That is also where the complexity lies. Most specialists master only one part of the puzzle: application development, automation, integration, or AI. But enterprise software requires all of these disciplines combined. It is precisely that cohesion that determines whether a solution remains scalable and manageable.

Several principles are essential in this approach. First: build the right thing. Do not start with building, start with understanding. Analyse the problem, the context, and the alternatives. Sometimes an existing solution is still the best option. In other cases, custom development makes all the difference.

Second: build the thing right. This means focusing on the foundation from day one: architecture, security, and scalability. AI can take over a significant amount of work, but it does not take over that responsibility.

Marijn van de Poel is Chief Proposition & Strategy Officer at Ciphix and has spent years helping organisations successfully embed technology into their operations. Interested or have questions? Feel free to get in touch via ciphix.io/contact