For decades, CIOs have faced a familiar decision when evaluating enterprise software: Should we build it ourselves or buy it from a software provider?
The answer has often favored buying. Commercial software, particularly software as a service (SaaS), provides mature capabilities, ongoing enhancements and technical support. Building custom applications typically requires significant investments in developers, infrastructure, testing, security and maintenance.
Artificial intelligence is beginning to challenge that equation. With AI-powered development tools, applications that once required weeks or months of programming can sometimes be prototyped in hours or days. Developers can generate code, build interfaces and automate testing with less manual effort. Even employees with limited programming experience can describe what they want an application to do and have AI generate a working solution.
This emerging approach, commonly called vibe coding, raises an interesting question for enterprise technology leaders: If AI makes software development faster and less expensive, should we reconsider how much we spend on SaaS applications?
I believe the answer is yes, but with an important qualification. AI is reducing the cost of creating software. It has not eliminated the cost of owning software. The opportunity for CIOs is to reconsider the traditional build-versus-buy decision while recognizing that enterprise software involves far more than generating code.
AI is changing the economics of software development
One of the primary reasons organizations embraced SaaS was the economics of software development. Rather than building and maintaining applications internally, companies could subscribe to established products and benefit from the vendor’s ongoing investments in functionality, security, infrastructure and support.
Over time, however, SaaS portfolios have become increasingly complex. Many enterprises now operate hundreds of applications, some with overlapping capabilities, limited adoption or subscription costs that continue to increase. Business units may purchase specialized tools to address relatively narrow requirements, creating additional integration and governance challenges.
AI-powered development introduces another option. Imagine a department paying an annual subscription for a relatively simple application that manages approvals, tracks requests or coordinates a specialized workflow. Historically, developing a replacement might have required a project team and several months of development. Today, AI-assisted development could make a custom solution economically attractive, particularly if the requirements are straightforward and the organization already has the necessary infrastructure and expertise.
The economics become more interesting when multiplied across an enterprise software portfolio. Replacing selected applications, eliminating overlapping functionality or building targeted capabilities around existing platforms could generate meaningful savings. However, the productivity benefits of AI-assisted development deserve careful examination.
A 2025 randomized controlled study by METR examined how AI tools affected experienced open-source developers working on familiar codebases. Surprisingly, researchers found that developers using AI tools took approximately 19% longer to complete their assigned tasks, despite expecting the tools to make them faster.
The study focused on a particular development environment and should not be interpreted as evidence that AI generally slows software development. It does, however, illustrate an important point: Generating code faster does not necessarily translate into delivering reliable software faster. CIOs need to evaluate actual development productivity rather than assume that AI automatically produces lower costs.
The hidden costs and risks of replacing SaaS
The excitement surrounding vibe coding is understandable. Watching AI generate a functional application from a few natural-language instructions can make traditional software development appear unnecessarily complicated. But enterprise applications have requirements that extend well beyond whether the software works during a demonstration.
Security is one of the most important considerations. A 2025 report from Veracode examining AI-generated code found that 45% of the code samples evaluated introduced security vulnerabilities. The research highlights why organizations cannot assume that AI-generated code follows established security practices, even when an application appears functional.
Enterprise applications require authentication, authorization, encryption, vulnerability management and data protection. Depending on their purpose, they may also need to satisfy regulatory, privacy and audit requirements. Those responsibilities remain with the enterprise regardless of how quickly the application was developed.
Maintenance presents another challenge. A SaaS subscription typically includes product enhancements, infrastructure management, patches and technical support. When an organization builds its own application, those responsibilities become internal obligations. Business requirements change, vulnerabilities emerge and integrations break. Applications must be monitored, tested and supported throughout their useful lives. AI may help perform these activities, but it does not eliminate the need for accountability.
This becomes particularly important when applications are created outside traditional IT development teams. Vibe coding can empower non-IT professionals to solve problems without waiting for formal development resources, especially when addressing simple departmental needs. But it can also create a new generation of shadow IT, where applications are deployed without sufficient oversight.
Consider what happens when an employee creates an application that becomes essential to a department’s operations. Who maintains it when the employee leaves? Who ensures that data remains protected and that the application continues functioning when underlying systems change? These questions become more consequential when applications support financial transactions, customer information or other business-critical processes.
Integration is another frequently underestimated cost. A seemingly simple SaaS application may connect to ERP, CRM, identity management and other systems. Replacing it may require rebuilding integrations, migrating data and ensuring that existing workflows continue functioning. An application that works well for a few users may also behave differently when supporting thousands of employees or large transaction volumes.
These considerations reinforce an important distinction: Vibe coding may make software easier to build. It doesn’t necessarily make software easier to own. For CIOs, the relevant measure is total cost of ownership, not simply initial development cost.
Rethinking the build-versus-buy decision
I do not believe AI-powered development means enterprises should begin replacing their major SaaS platforms indiscriminately. Mature ERP, HR, payroll and financial applications contain years of accumulated functionality, business rules and compliance capabilities. Recreating those capabilities internally would introduce significant complexity and risk, even with improvements in development productivity.
The more immediate opportunity may lie in specialized SaaS applications that address relatively narrow, non-critical requirements. Some duplicate capabilities already available in enterprise platforms. Others provide functionality that could potentially be developed internally at a lower long-term cost. These applications deserve another look.
Rather than applying the same build-versus-buy criteria used five years ago, CIOs should evaluate their portfolios through a new lens. I would consider five questions when making these decisions.
- Is the capability strategically differentiating? Custom development may provide greater flexibility and control when an application supports a unique business process. Standardized capabilities that offer little differentiation may be better served by commercial products.
- How complex is the application? Simple workflows, reporting tools and departmental applications may be good candidates for AI-assisted development. Applications involving complex transactions, regulatory requirements or mission-critical operations require greater scrutiny.
- What is the true total cost of ownership? The analysis should include subscription fees, development, infrastructure, integration, security, maintenance and ongoing support. Internal resources required to operate the application must also be considered.
- Do we have the capabilities to support what we build? AI may reduce manual programming, but organizations still need architecture standards, secure development practices, testing and operational support. Without these capabilities, technical debt can accumulate quickly.
- Can we achieve the same outcome by extending what we already own? Before replacing a SaaS application, organizations should examine whether existing enterprise platforms can provide the required functionality. Many companies are already paying for capabilities they have not fully adopted.
The best alternative to purchasing another SaaS application may not be building a new one. It may be making better use of technology already available.
CIOs should also distinguish between AI-assisted professional development and vibe coding by business users. Professional developers can use AI while applying established engineering practices. Non-IT professionals (AKA users) can create lightweight applications within defined boundaries. Both approaches have value, but they require different levels of governance.
There is another implication worth considering. The same AI capabilities that make custom development more attractive are also available to SaaS providers. Vendors can use AI to accelerate development, automate support and introduce new functionality. As development economics change, CIOs should question whether traditional SaaS pricing models remain appropriate.
If AI reduces the cost of developing and maintaining software, will those efficiencies eventually be reflected in subscription prices? Or will vendors continue increasing prices as they introduce AI features and new licensing models? These questions are worth raising during software renewals and commercial negotiations.
In many cases, the best answer will be a combination of building and buying. Organizations may retain core enterprise platforms while developing targeted applications, workflows and AI agents that extend their functionality. This approach can preserve the stability of established systems while allowing businesses to innovate more quickly.
AI is creating opportunities to simplify application portfolios, reduce unnecessary software spending and develop capabilities that better serve the business. Realizing those benefits requires disciplined evaluation rather than enthusiasm for how quickly software can be generated.
The question is no longer simply whether we can build an application more cheaply than we can buy it. It is whether building, buying or extending existing technology will deliver the best business outcome over the application’s entire life.
That is the decision CIOs need to get right.