Vibe coding and shadow IT: how can companies discover and govern homegrown applications?

A steel chain with its middle link replaced by a hand-shaped red plastic link.

One colleague uses AI to put together a quote tracker. Another automates invoice processing. An assistant builds a simple CRM because the shared spreadsheet has become unmanageable. The solutions work, so they soon hold customer data, internal documents and business decisions. Yet IT, the person responsible for data protection and management may not even know they exist.

The vibe coding shadow IT problem starts here: development becomes easier, but accountability for applications does not automatically fall into place. In this instalment of “The 10 biggest risks of vibe coding”, we explore how to discover and govern homegrown applications. One particularly important question is what happens when the employee who created software that has become essential to business operations leaves.

1. When does a good initiative become shadow IT?

Shadow IT means using IT tools and solutions that operate outside an organisation’s approved IT governance framework. It does not require malicious intent. Often, a helpful employee is simply trying to solve a longstanding problem.

Vibe coding accelerates this: natural-language instructions allow even someone unable to verify the security of generated code to build a working application. A working interface, however, is not proof of secure operation. Just because a CRM lets you add customers does not mean its access management, backups or data deletion are adequate.

The turning point comes when an experiment becomes a live business tool. If colleagues use it, it stores personal data, connects to a company system or informs decisions, it can no longer be treated as a personal project. A manager having seen and praised it does not make it compliant either.

2. Discovery: start with an application inventory, not a prohibition notice

A blanket ban can easily drive homegrown solutions further underground. A better approach is to start with a disclosure period: the aim is to address risks, not automatically penalise previous initiatives. Without management support, employees may easily see this as yet another administrative burden.

The survey should not simply ask, “Have you developed an application?” Many people do not consider an AI-generated web form, a browser-based customer list or an automated data-copying script to be an application. Also ask: without which of your own tools would your work stop or slow down?

  • Business purpose: what task does it support, who uses it, and how essential is it?

  • Data: what personal, confidential or financial information does it handle?

  • Technical setup: where does it run, where is the code stored, and which services does it connect to?

  • Access: who is the administrator, and who owns the hosting account and subscription?

  • Accountability: who is the business owner, who fixes it, and who takes over if they are unavailable?

  • Rights: what assigned work duties, agreements and third-party licences apply to it?

Supplement disclosures with targeted checks of subscriptions, cloud accounts, code repositories and system integrations. However, employee devices or communications may only be inspected with an appropriate legal basis, proper notice and in a proportionate manner: discovering shadow IT does not override data protection.

Colleagues review a work laptop together while one takes notes; the screen content is not readable.

3. Risk classification: do not apply the same rules to every application

A demonstration using test data and a CRM managing the entire customer database do not pose the same risk. Classification depends on data sensitivity, accessibility, associated permissions and the business impact of downtime. For tools already in production, a decision is also needed on the conditions under which they may be used until the review is complete. The risks may justify restricting or suspending use; in such cases, temporary arrangements for carrying out the affected business task must also be planned.

  • Low risk: an isolated experiment, with no live data or business dependency. Registration and basic usage rules are required.

  • Medium risk: a tool used by an internal team, with a limited scope of data. A designated owner, access management, tested backups and a technical review are warranted.

  • High risk: customer data, sensitive information, internet access, elevated system privileges or a critical business role. Documented security approval and, where necessary, legal and data protection approval are required before production use.

The security review should examine authentication, server-side authorisation, secrets storage, logging, dependencies and recoverability. A typical risk arises when the interface hides other customers’ data, but the API returns it without proper checks.

Customer data or trade secrets pasted into an AI tool constitute a separate data flow. The provider’s terms, data retention, purposes of use and any international data transfers must be checked. For personal data, the legal basis under the GDPR, transparency, a data processing agreement and security are also relevant. A data protection impact assessment is required when the processing is likely to result in a high risk, not simply because AI assisted with development.

4. Who owns the CRM if the employee who built it resigns?

Suppose an assistant builds a CRM on their own initiative. Development is not part of their job description, yet the system comes into daily use. When leaving, they state that the software belongs to them and demand payment for its continued use. Neither the validity of that demand nor the company’s automatic acquisition of all rights can be taken for granted.

The job description is important, but it is not the only evidence

Under Hungarian copyright law, a key factor for works created in the course of employment is whether creating them was part of the employee’s employment duties. Unless otherwise agreed, the employer acquires the economic rights upon delivery of such a work. However, the employment contract, assigned tasks, instructions and actual circumstances must all be considered together.

The absence of “software development” from the job description does not, on its own, settle the dispute. Likewise, using a company laptop, working during business hours or using the software for company purposes does not prove that rights were acquired automatically. If the development genuinely fell outside the employee’s duties, a separate licensing or rights assignment agreement may be necessary.

Code, data and access are separate issues

Copyright protection for AI-generated code also depends on the human creative contribution. The entire program cannot automatically be treated as the exclusive creation of the employee who entered the prompt. External libraries, reused code snippets and service provider terms may also affect how it can be used.

Any rights relating to the software do not give unrestricted control over customer data or trade secrets. Taking them, deleting them without authorization or making them inaccessible may constitute separate legal violations. Meanwhile, code stored in a personal account can create an operational dependency even if the company otherwise has a strong legal position.

This is general information, not legal advice on a specific case. Before pursuing a particular dispute, it is advisable to have the documents reviewed by employment law and copyright specialists.

A dark technical diagram with separate code, data and access channels, featuring legal, data protection and handover checkpoints.

5. Approval and offboarding: govern the entire lifecycle

A sustainable solution is not a single approval email, but a traceable process. Establish a separate experimental environment, clear go-live criteria and a designated decision-maker. Approval should apply to the specific version, data scope and intended use; a new integration may require a new review.

  1. Idea submission: document the purpose, expected benefits, data and planned services.

  2. Preliminary classification: designate a business owner and technical lead, and define review requirements.

  3. Clarifying rights: document task assignments, necessary agreements and licenses.

  4. Go-live gate: verify testing, access permissions, backups, cover arrangements and operating documentation.

  5. Periodic review: assess changes, permissions, vulnerabilities and the continued need for the application.

When an employee leaves, use a handover checklist covering the code repository, hosting, domains, subscriptions, integrations and recovery procedures. Revoke permissions, rotate the necessary secrets and test the replacement’s access. Rights and handover arrangements must be settled before go-live, not on the day someone resigns.

6. xFLOWer: processes managed within a shared framework instead of homegrown applications

Many homegrown applications actually address a process need: someone wants to assign tasks, request approval, track deadlines or handle customer cases. In such situations, first consider whether a new, standalone application is needed at all.

An xFLOWer-based approach focuses on processes managed within a shared framework. The assessment can clarify how roles, tasks, approvals and system connections can fit together, and where custom development remains justified. This allows the employee’s business knowledge to be put to use in transferable operating procedures rather than an application tied to one person.

A workflow framework such as xFLOWer can help reduce shadow IT when both the shared platform and its custom connections operate under controlled conditions. If the business need can be met on the platform, custom development may be limited to interfaces connecting existing systems. These still require an inventory, a designated owner, access reviews, testing and approval: connecting to a shared platform does not, in itself, make a custom solution compliant.

Within a shared framework, prototyping, testing and production use can all be kept under control. This requires appropriate permissions, logging and retention rules so that data exports, data transfers and scheduled deletions can also be tracked.

A split image: on the left, a desk cluttered with papers; on the right, an organized workstation with a simple flowchart featuring checkpoints.

Customizing existing software can reduce the legal uncertainty associated with developing an in-house application, but rights to custom code and customizations must still be clarified.

This is no substitute for security planning and clarifying legal rights. The aim is to reduce the number of isolated tools that need to be maintained separately and make responsibilities clear.

In summary: the goal is not to eliminate employee initiative, but uncontrolled operations. Start by assigning someone to compile an application inventory and ask teams to report the tools they have created for their daily work. After classifying them, decide whether to retain, improve, replace or retire each one. A shared framework such as xFLOWer may also offer a solution, but responsibilities, rights and handover arrangements must be clarified in every case.

Want to uncover homegrown applications across your company? Consult the xFLOWer team about which solutions are worth retaining and governing, and which business needs would be better addressed within a shared process management framework.

This article is part of our complete guide: The top 10 risks of vibe coding: a complete enterprise guide from security to copyright

An unlabeled editorial infographic with a pie chart, bars and a curve, featuring gray and red highlights.

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.

Contact us

Let's talk about your processes

Tell us briefly which process you'd like to automate – we'll get back to you within 1 business day.