Fear of SaaSpocalypse has swept the land as a handful of high-flying startups have very publicly ditched big systems of record for internal solutions. Their rationale? Making bespoke software is now cheaper than paying for SaaS.
Make no mistake: the ability to build an application by describing what you want is a significant change in who can create software and what it costs. In our daily practice, we are not only witnessing but enabling more structured forms of vibe coding and other DIY technology creation. But the specific application someone creates is often only one part of a much larger business process, and its usefulness depends on connections to systems that already exist.
This fact is playing out among CIOs.
In Kyndryl’s 2026 modernization research, AI is leading to massive buildouts and investment across the enterprise. Sixty-eight percent of organizations said they plan to increase their use of at least four technology categories, and half plan to increase their use of five or more. Across every category measured — SaaS, public and private cloud, mainframe, edge computing, sovereign cloud zones and on-premises infrastructure — more organizations expect usage to increase than decrease. Read it again: that includes even so-called legacy technologies, and newer ones. And 77% plan to increase SaaS investments over the next year; just 7% think they’ll decrease them.
Vibe coding expands the technology stack
Consider a manufacturing supervisor who wants to understand why particular orders keep missing their delivery dates. With AI assistance, they might build a tool that brings together production schedules, maintenance records, and supplier information, then highlights patterns that were difficult to see when those records sat in separate applications.
They might even train a bespoke small model using historic and PLC data, customized to the organization’s production lines, equipment and firmware. The code and system they create build additional functionality. This factory expert understands the problem because they deal with it every day, and now they have a more accessible way to work on a solution.
But to be useful, that tool still needs the components of its integral sum, such as the systems that record orders, manage inventory and schedule production. It needs accurate information about customers and suppliers, and it needs to respect the permissions governing who can see or change that information.
In other words, building the application may have become much easier — or even imaginable for the first time. But the underlying systems not only continue to remain essential but also may need to expand to accommodate the massive new data requirements unlocked by the bespoke systems that the factory supervisor has built.
As AI has reduced the resources required to create these bespoke products, more of these projects can coexist while still adding value to the enterprise. This is a classic manifestation of Jevons’ paradox, an economic principle that describes how greater efficiency can lead to greater total consumption when lower costs encourage more use.
By extension, Jevons’ paradox also implicates expansion in any resource class required to maintain the apex resource — in our case, software, which increasingly equates to AI. Our research shows that businesses expect to use a wider range of technologies as AI changes their requirements. AI workloads, for instance, are a top driver of increased edge computing usage, and new AI capabilities on the mainframe are the top reason for increased usage of that platform. The latter is an example of usage expanding on a platform that some people might assume would be displaced by newer technology.
More software creation means more human responsibility
The bottleneck traditionally has been this: more people understand a business problem than can build software to address it. A claims specialist knows which exceptions slow down a case. A warehouse manager knows where information gets lost between shifts. A finance colleague knows which reconciliations consume days of repetitive effort. Giving these people more ability to create tools could allow companies to address work that has remained on the waiting list for years.
Employers must teach them how to do that responsibly. Explaining a requirement to an AI system is a useful starting point, but employees also need to test whether the result is correct, understand which data it uses, and recognize when a proposed change needs specialist review. The business knowledge that makes someone a good creator does not automatically give them the skills to operate an application safely.
AI will build upon the existing technology foundation—with more applications, more people participating in their creation, and a greater need to connect technical choices to business consequences.
Build vs buy: how IT teams will make the hybrid software future possible
Technology teams will have an important role in making this practical. They can provide approved connections to company systems, reusable components, testing support and a clear route for an employee’s useful experiment to become a maintained business application. They will also need to decide who owns a tool when its creator changes roles and how to retire it when the business no longer needs it. Or, they can decide that existing systems are a better fit and should be customized.
Of course, the old work of cataloging and integrating remains both pressing and unfinished. Kyndryl’s research finds that 51% of respondents identify integrating legacy systems with the rest of their estate as the greatest obstacle to modernization. Adding more applications without addressing those connections could make an existing problem worse. The ability to create software quickly therefore makes integration, maintenance and clear ownership more important parts of the picture.
However, it is a grave mistake to build a workforce and technology strategy on the assumption that easier software development makes the underlying systems unnecessary. Rather, AI will build upon that substrate, with more applications, more people participating in their creation, and a greater need for colleagues who can connect technical choices to business consequences.
I encourage leaders to ask their employees which useful tools they have never been able to get built, then consider what would be required to create and support them. Now is the time to rethink your IT estate and build and orchestrate for the AI future.
