Vibe Coding vs. Enterprise Software

In the first quarter of 2026, the stock prices of many SaaS providers plummeted. The declines ranged from 20% to as much as 70% for established industry leaders such as Salesforce, Atlassian, Figma, and Workday. One of the factors influencing investor sentiment was the growing capabilities of tools such as Claude Code, as well as concerns that AI would allow customers to independently create software perfectly tailored to their needs. The scale of this phenomenon was so significant that the media began referring to it as the “SaaSpocalypse.”

Artificial intelligence actually makes it possible to create extremely impressive prototypes of business applications in a very short amount of time. For example, we can import a company’s financial report (downloaded from an ERP system as an .xls file) into a small local program generated by AI and, in just a few moments, obtain a visualization of the data and a handful of insights.

Various forms, dashboards, and simple databases are created by AI agents in a matter of days or hours, rather than the months-long projects of the past. What’s more, they can be created by people who don’t know any programming languages or the principles of software engineering. This method of application development is often referred to as “vibe coding” (from the English word “vibe,” which means, among other things, following one’s intuition; we wrote about this in more detail in the article titled “Will AI Replace Programmers?”).

Such a revolution in app development (and the media hype surrounding it) may convince management that the company has become self-sufficient in software development. However, while analyzing a financial report or a simple form are valuable applications of AI, they are still very limited in scope. AI products do not constitute robust and comprehensive IT systems for businesses.

Let’s consider a seemingly simple example: we want to streamline the process of submitting leave requests and create a clear way to display the absence schedule so that every employee can easily see who is available and when. Of course, if we enter such a prompt into one of the AI tools, we’ll end up with an application—and a visually appealing one at that. However, when we try to actually implement it in the company, we’ll run into several important questions:

• Where should the system retrieve the current list of employees and information about reporting relationships within the organizational structure (so that a supervisor can approve leave requests from their subordinates)?
• How can I integrate the app with an email system to send notifications about pending tasks (application approvals) to the appropriate recipients?
• Can we guarantee that the system will correctly determine the available types of absences (in accordance with the Labor Code) and the number of days allowed for absences for each person in the current year (e.g., based on length of service)?
• How can information be made available online in a way that ensures it is not accessible to unauthorized users?

Of course, the issues mentioned above are easy to resolve in a company with a few or a dozen or so employees. But in such an organization, a shared file (e.g., a Google Sheets spreadsheet) is all we need to plan vacations, rather than dedicated applications. Meanwhile, the most optimistic AI enthusiasts claim that it is ready to replace entire ERP or CRM systems.

A stable and reliable corporate IT environment requires:

• Access management: SSO, permission control, revocation of access upon termination of the contract, comprehensive reporting for audit purposes;
• Cybersecurity and auditability: protection against data leaks, resilience to attacks, and detailed logs (who performed which operation and when);
• Data architectures: a consistent interpretation of concepts across the entire company (e.g., “Do ‘employees’ in the report refer to the number of people or the total number of full-time equivalents?” and hundreds of similar decisions/rules);
• Performance and scalability: well-designed data replication, proper database structure, query optimization;
• Process logic: workflow mechanisms and operating rules aligned with the company’s structure (e.g., “forward the request to the supervisor according to the organizational structure, and if the amount exceeds 1,000 EUR, forward it to the supervisor two levels up”).

These issues will not be properly addressed at the organizational level by users who “code on a whim” to create their own micro-tools—no matter how many millions of tokens they use or how many times a chatbot praises their ingenuity. This approach will not result in secure databases or systems that coordinate the work of hundreds of employees according to documented business rules.

Does this mean that users creating apps with AI is just a passing fad? Quite the contrary. The key to success is placing this grassroots development within the right framework and integrating it with traditional software.

We should not look for shortcuts or bypass the implementation of key systems such as ERP, CRM, or document workflow platforms. They are an essential foundation for a secure architecture. However, we should expect their providers to offer flexible application programming interfaces (APIs). Modern AI tools excel at analyzing APIs and allow us to extend the functionality of SaaS systems (as well as more traditional on-premises solutions) so that they better meet users’ needs.

The APIs provided by system vendors should meet the following conditions:

• Data Retrieval: APIs should only provide access to the same scope of information that a user would have access to after logging into the main system (e.g., ERP). AI can then visualize this data in any way that is useful to users—without involving the company’s IT department or software vendor.
• Data modification: This is a more sensitive area, but a properly defined API will verify permissions and ensure data consistency (for example, business rule validation will prevent the entry of a transaction with an amount exceeding the limit for a given position).

It’s worth exploring the possibilities of expanding your existing systems through APIs in combination with AI-driven application development. If your current software has too many limitations in this regard, now is the right time to consider an upgrade (e.g., implementing solutions that support such an architecture—such as the Rockawork process digitization platform or the BeeOffice employee self-service system).

Thanks to AI-driven development of new features, modern SaaS systems are becoming even more useful and helping to boost business productivity. Users, in turn, are more satisfied because they gain tools that are perfectly tailored to their everyday needs.