How AI gives you back control over your software and your profits

AI driven software engineering frees companies from the constraints of SaaS software licences, but only with the right approach.

Jan Baan and Ad Nederlof say out loud in their article in the dutch paper FD what the IT sector has known for years: SaaS software companies systematically skim off their customers’ productivity gains. AI offers a way out, but Vibe Coding is not the answer.

The heart of the problem

SaaS companies have built a business model around lock-in, not delivered value. The figures speak for themselves. On average, an organisation uses only 40% of the available functionality. Yet it pays 100% of the licence costs year after year. On top of a majority of the employees who have a licence, barely use the software. And almost every organisation eventually builds its own additional solutions. In the past, that might have been an Excel macro. Today, it is low-code tools and, increasingly, Vibe Coding. We do this because SaaS software never quite covers the unique business process precisely enough.

Do not get me wrong: these software packages deliver significant productivity gains for the companies that use them. However, there is also a downside. The exit threshold is deliberately kept high through complex migrations and complicated configurations. This gives suppliers room to raise annual licence prices, knowing that customers will not leave quickly. As a result, much of the productivity gain created by the software flows to the SaaS supplier rather than to the customer’s bottom line.

AI-generated software as a lever, but not in the way the hype suggests

The term SaaSpocalypse has been circulating in the sector for some time. The announcement of a new AI model from Perplexity recently led to a fall in the share prices of several SaaS suppliers. That shows how seriously the market is taking this risk.

But the conclusion that Vibe Coding, where AI generates software based on broad functional instructions, is the solution is wrong. Vibe Coding produces quick results, but not manageable or scalable software. Certainly not for critical business processes.

Spec-driven AI engineering: software that belongs to you

The real shift comes from a different approach: spec-driven AI engineering. In this model, AI agents drive software development, but always within strict frameworks that apply to your organisation. Think of precise specifications, architectural guidelines, clear quality standards, full traceability, efficient use of CPU capacity and AI tokens, and zero trust principles. The AI works in a sandbox with clear boundaries, not in an uncontrolled way.

The quality of the result depends on three elements. First, a razor-sharp scope: the agent knows exactly what it must and must not do, leaving no room for interpretation. Second, mature skills: defined and tested capabilities that the agent can apply. This is not ad hoc generation, but reproducible craftsmanship. Third, a clear definition of tools: every tool the agent uses is explicitly specified, with clear inputs, outputs, and boundaries.

Organisations that put these three elements in place can build software at a level that Vibe Coding simply cannot match. The argument is not just that it is faster or cheaper, but that it is demonstrably better, more manageable, and more closely aligned with the business process.

Demonstrably compliant by design

This is where the debate about AI-generated software misses a dimension that is decisive for many sectors: compliance.

Software built through spec-driven AI engineering is fully and demonstrably compliant. Every design choice can be traced back to a specification. Every function can be traced back to a requirement. The entire architecture is documented from the frameworks within which the agents operated. This makes it possible to show a regulator or auditor, whether in the context of ISO 27001, NEN 7510, DORA or any other certification, exactly what the software does, why it does it and how it does it.

This is not something you add afterwards to code generated through Vibe Coding. In spec-driven engineering, compliance is built into the process.

For organisations in finance, healthcare, government, or critical infrastructure, this is not a side issue. It is the condition under which software is allowed to go into production at all. And it is precisely the reason why Vibe Coding is not a serious option in those sectors, however attractive its speed may appear.

The nuance is missing from the debate

This does not mean that SaaS will disappear. For commodity processes such as HR administration, email, and basic CRM configuration, standard software remains effective. The disruption will mainly affect the layer where companies currently pay the most for the lowest return: specialist functionality that never quite fits, expensive seat licences for occasional use, and integrations that require more custom development costs than the SaaS solution itself.

That is exactly the layer that can be replaced with spec-driven AI engineering by software that does fit completely, and where the productivity gains from that software are no longer skimmed off by a supplier with a structural information advantage.

Conclusion

SaaS applications often fail to meet functional requirements and are used only to a limited extent. The remedy, however, is not uncontrolled AI code generation, but a structured approach in which AI serves as an engineering tool within strict frameworks. Spec-driven, zero trust, and fully owned by the customer. That is how you remove the constraints of SaaS without introducing new risks.