An employee builds a CRM with AI assistance over a few afternoons. Customer registration works, reminders are generated, and the manager can view the sales list. Why would a longer development project be necessary? The answer often emerges only later: when customer data leaks, the application goes down, or the only person who knows how it works leaves.
The vibe coding or low-code question is therefore not merely a technology choice. We are also deciding who controls access, who is entitled to use the software, and who keeps the process running. As part of our coverage of the “10 biggest risks of vibe coding,” we now examine the business, security and legal aspects of this choice.
1. Three approaches, different models of responsibility
Vibe coding here means having software generated through natural-language instructions and accepting the result primarily based on its visible functionality. The risk lies not in using AI itself, but in failing to have the code, architecture and security professionally reviewed.
The low-code approach builds applications from predefined components through configuration and, where necessary, supplementary code. It can reduce the amount of custom code required, but does not guarantee security in itself. The platform’s capabilities, settings and operating conditions still need to be assessed.
The custom development approach offers greater flexibility for specialized business logic and integrations. At the same time, the client and developer must organize testing, documentation, vulnerability management and maintenance.
These three paths are not mutually exclusive: custom development can also be AI-assisted, and a low-code system can incorporate custom components. The dividing line is control, not the presence of AI.
2. Shadow IT: when a prototype quietly becomes a production system
Shadow IT encompasses applications and services used outside IT oversight. Vibe coding makes creating such solutions significantly easier: business teams no longer necessarily need to find a developer to launch their own system.
The typical problem does not arise from malicious intent. A helpful colleague solves a local issue, and others then start using the application. A personal account becomes corporate infrastructure, and a test database becomes a customer registry, without formal approval.
Data processing blind spot: it is unclear which data ends up with which service providers, and under what terms.
Access control failure: login works, but the server does not properly check who is allowed to view each customer’s records.
Secrets management issue: an API key or database password may end up in source code, in the browser or in an AI conversation.
Operational gap: there is no tested recovery procedure, update schedule or designated incident owner.
From a GDPR perspective, the purpose and legal basis of processing, relationships with data processors, retention and any international data transfers must all be clarified. An employee’s personal initiative does not exempt the company from its obligations as a data controller.
The solution is not necessarily a blanket ban. An approved sandbox, prototyping without live data and a clear go-live approval gate may be more useful.
With a workflow framework under corporate oversight, such as xFLOWer, custom development may be needed only for a new interface connecting to existing systems, rather than for a standalone shadow system. This still requires review, but the process can remain within a shared access control and operational framework. From a data protection perspective, the aim is to control and log deletions, data transfers and exports, and to track retention deadlines. This requires configurations and checks that also cover connected systems: the framework alone does not automatically ensure GDPR compliance.
3. If the secretary who built the CRM resigns, who owns the software?
The secretary in this example builds a CRM, even though programming is not part of their job description. On leaving, can they take the system with them, prohibit its use or demand payment for it? This cannot be determined solely on the basis of a single sentence in a job description.
Actual employment duties also matter
Under the general rule of Hungarian copyright law, unless otherwise agreed, the employer acquires the economic rights to a work created in fulfillment of employment duties upon its delivery. Special rules also apply to software. The employment contract, job responsibilities, employer instructions and circumstances of creation must be examined.
Using a company laptop and working during business hours do not necessarily determine the legal position on their own. Equally, the absence of the term “software development” from the documentation does not automatically prove that the employee holds exclusive rights.
The software, the data and access are separate issues
If development was not part of the employee’s duties and no appropriate agreement exists, the company’s right to use the software may become uncertain. The rights holder of copyright-protected elements may negotiate licensing and compensation. This does not, however, mean they can lawfully delete company data, take the customer database or paralyze operations.
With AI-generated code, a further question is which elements involve human creative activity sufficient to establish copyright protection. The AI provider’s terms and the licenses of incorporated third-party components must also be checked. It is unsafe to assume that “AI wrote it, so it is free to use.”
Prevention requires written agreements on rights and a structured handover: a company code repository, company administrator accounts, documentation and operational arrangements that allow others to take over. In a specific dispute, consulting employment law and copyright specialists is advisable.
The situation may be simpler within a workflow framework: the company's contract governs use of the underlying software, while the employee customizes processes in an existing system. Operations therefore do not depend on the legal status of an entire application created by that employee. Even so, usage rights for custom extensions and process models, along with data access and exportability, should be agreed in writing.
4. Vibe coding or low-code? A decision matrix based on five criteria
Do not decide solely on development speed. For each of the five criteria below, first document your own process requirements, then indicate which solution meets them and under what conditions. Faster delivery does not compensate for unmet mandatory security and operational requirements. If several approaches are suitable, narrow down the options based on customizability, total lifecycle cost and ease of handover.
Business criticality. Vibe coding: for isolated prototypes with limited consequences if things go wrong. Low-code: for recurring business processes, with appropriate operational guarantees. Custom development: for specialized or critical operations, provided the necessary engineering and operational support can be secured.
Data sensitivity and control. Vibe coding: initially with synthetic test data. Low-code: a good candidate if it demonstrably supports the required access management, logging and data handling conditions. Custom development: for specialized control requirements, with dedicated security design and verification.
Customization. Vibe coding: for quickly testing ideas. Low-code: for tasks, approvals and status transitions that can be readily modeled. Custom development: for unusual algorithms, performance requirements or logic beyond the platform's limits.
Integration and lifecycle. Vibe coding: for simple, self-contained experiments. Low-code: if the necessary connections and version control capabilities are available. Custom development: for complex integrations, with a commitment to covering maintenance costs.
Sustainability and exit. All three require clear ownership, documentation and access arrangements. For low-code, specifically assess exportability and vendor lock-in; for custom and AI-generated code, assess ease of handover and dependency management.
Disqualifying condition: a production system handling sensitive data must not go live without verified access management, clearly defined data handling arrangements and a designated operations owner, regardless of the technology used.
Estimate costs over the full lifecycle for all three approaches: alongside development or configuration, account for integration, testing, licenses, operations, support, maintenance and the cost of future handover or replacement. Include token costs for any AI usage in the estimate as well. Workflow solutions may offer the advantages of building on existing components and greater predictability, but these should be compared with the other two approaches based on specific requirements and fees.
5. xFLOWer: when is AI-assisted process design an advantage?
For xFLOWer, the starting point is the business process: who initiates it, who approves it, which rules govern status changes and what happens in exceptional cases. AI delivers business value when it accelerates the design of this process rather than bypassing established responsibilities.
A workflow framework such as xFLOWer can offer an alternative to uncontrolled application development: the business process is modeled within defined boundaries, rather than building new software foundations every time. This offers a security advantage if the chosen framework is audited, tested and subject to professional oversight independent of the process creator. The audit's scope and currency must be verified, and new processes, permissions and integrations must also be tested.
When evaluating the capabilities of xFLOWer and xFLOWer.ai, it is therefore worth working with a specific list of requirements. During the demonstration, do not just ask to see a process being created; also examine how permissions, changes, errors and handover scenarios are handled.
An approval chain, contract preparation or internal request management provides a particularly useful evaluation scenario. In these cases, controlled steps and responsibilities may matter more than unrestricted programmability. However, always verify the actual capabilities and contractual terms of the chosen solution.
6. From a working prototype to a dependable system
If a homegrown CRM already exists, do not automatically delete it or accept it without review. First assess its data, dependencies, access arrangements and usage rights. Then decide: can it be made secure, does it need rebuilding or should it be replaced?
Assign business and technical owners.
Bring accounts, code repositories and subscriptions under company control.
Define rights and handover obligations in writing.
Test permissions, backups and the ability of others to take over.
In a workflow system, process models can also have a controlled lifecycle: demonstration prototype, testing, training environment and then approved production operation. During selection, ask for a demonstration of how these stages are separated and how handover gates work. This allows even the prototype to be built within a controlled framework, rather than having to establish secure operations around it later.
In summary: vibe coding may be suitable for quickly testing ideas without production data; a low-code platform that meets the requirements may suit readily modeled business processes; and custom development may be appropriate for logic or integrations that exceed platform limits. These approaches can also be combined. Base the decision on verified compliance with requirements, not on an impressive prototype: production use always requires security controls, clearly established usage rights and a designated operations owner.
With a properly selected and supervised workflow framework, such as xFLOWer, rapid process design can be governed by shared security, data handling and lifecycle rules, under more predictable operating conditions.
Which approach fits your process? Discuss the capabilities of xFLOWer and xFLOWer.ai with the Click On team, and start with a requirements-based review of a specific process.
This article is part of our comprehensive guide: The Top 10 Risks of Vibe Coding: A Comprehensive Enterprise Guide from Security to Copyright
xFLOWer.ai
Design the process behind this article – for free!
AI reads the processes hidden in this article; open any of them in xFLOWer.ai with one click.
