The emerging practice of ”vibe coding,” where developers rely on AI-generated code based on intuitive or aesthetic judgments rather than rigorous technical understanding raises a series of ethical concerns in regards to risks of security vulnerabilities, technical debt, and unclear responsibility allocation. Beyond these issues, the paper considers the potential for both technical and moral deskilling, as reliance on AI may erode programming expertise and diminish developers' ethical agency. To address these risks, the paper explores the introduction of purposeful friction in development workflows, understood as interventions that seek to foster human agency and critical engagement. Ultimately, this paper proposes that purposeful friction might have the potential to balance the advantages of vibe coding with the need to preserve responsibility, skill formation, and sustainable development practices.
Over the past few years, AI-assisted programming has shifted from experimental novelty to everyday
practice. Tools like GitHub Copilot that ofer explanations and generate software code that can be
readily integrated in software projects are being integrated into development workflows at an increasing
pace. This transformation is driven by the promises of increased productivity and lowered barriers
to entry [
Broadly speaking, vibe coding defined the use of an AI assistant to generate code while the person guides the process with ideas and descriptions in the form of natural language prompts. It distinguishes itself from earlier forms of AI-assisted coding in that in the latter, the AI assistant generates suggestions or blocks of code for the task at hand, and it might help identifying bugs and suggesting fixes, but the programmer needs to ultimately revise and accepts the suggestions and is the one actively debugging the code. On the other hand, in vibe coding, if there is an error, the coder is meant to merely copy and paste the error as it is generated back to the AI assistant for resolution, and any corrections should be accepted without checking.
While this approach has been justified as allowing the coder to express creativity in a way that formal programming does not, it also introduces a further level of ethical and technical detachment in the field that raises ethical challenges like security vulnerabilities in commercially deployed products, technical debt, difused allocation of responsibility, and the potential of technical and moral deskilling
CEUR
ceur-ws.org of programmers that emerged with the introduction of generative AI. This paper explores these ethical concerns, and it seeks to argue that the risk of technical and ethical deskilling from vibe coding could be mitigated with the implementation of purposeful friction.
Digital environments are often designed for seamlessness and flow. A general argument in favor of
AI-assisted programming is the ability of the tools to anticipate the needs of the developer, provide code
snippets that can be easily integrated into the code base, and enable the assembly of working prototypes
with minimal technical efort. The emergence vibe coding, on the one hand, follows the known focus
on providing seamless user experiences and quick results, and is viewed in a largely positive light in
corporate and entrepreneurial spaces [
From a purely technical standpoint, vibe coding became possible because AI assistants have reached
a level of technical sophistication that allows people without programming expertise to describe and
iterate ideas either through written prompts or speech recognition tools. Although initially, the primary
usage of vibe coding circulated around developing quick prototypes and kick-starting personal projects,
in the last months the rise of AI development platforms that employ vibe coding suggests that companies,
startups and programmers are leaning towards using vibe coding to launch products to the market
1. The why or how the generated code works loses relevance in the face of faster results and a lower
entry-barrier for non-programmers. However, vibe coding is also the consequence of deliberate choices
by field experts of how the process of developing software should be and about the future of the
technology industry. In a world that has embraced the implementation of AI in many areas of human
activity, and in particular with the rush of established and new generative AI models to take the lead
in the race almost every week [
However, there are epistemic and ethical costs associated with these goals [
Closely interconnected, a second ethical implication of vibe coding is technical debt. This a well known
phenomenon in software engineering and it is broadly defined as the accumulation of consequences
of decisions made during development, implementation or maintenance that prioritized immediate
1Consider the emergence of tools like Lovable, Cursor, Replit and more recently, Google’s Opal
goals over quality 2. It has been argued that technical debt can negatively afect productivity, long-term
product sustainability and team morale, and it is also associated with real monetary consequences [
The third ethical issue is difused responsibility allocation within technology organizations. It has
been established that we are in the middle of a Fourth Industrial Revolution characterized by the
radical transformation of virtually all of human activity spheres by the convergence of technologies
like AI, robotics and gene editing [
Another, more subtle risk is the potential of vibe coding to lead to a gradual deskilling of developers.
This phenomenon, as argued elsewhere [
Broadly construed, deskilling is the situation within a certain occupation in which the workers
experience reduced “discretion, autonomy, decision-making quality, and knowledge as they perform
their jobs. All of which negatively afects their ability to perform these jobs in a manner they see fit”
[
Vibe coding risks to diminish the value of the technical judgment employed by developers during the
ideation and development phases of a pipeline and the critical evaluation of the potential outcomes of
decisions regarding the code’s architecture and functioning once deployed to the consumers. Instead
of cultivating a deep understanding of code structures and system behaviors, users may come to rely
solely on afective cues, i.e., what “looks right,” “sounds right,” or “feels like it works.” This implies
that much of the cognitive value of software development is being replaced by simple prompt-based
interactions and automated model inference, thus a potential downside of vibe coding is the loss of
the skill formation that comes with grappling with the ”material resistances inherent in code’s syntax
structure and debugging challenges in a manual workflow” [
In regard to moral deskilling, vibe coding reinforces a long-standing concern in human-computer
interaction about the automation of expertise [
As Grifin and colleagues have argued, the matter of the moral agency of programmers is an oftenoverlooked aspect of software development [ 17]. While in many other fields it is required of professionals to comply with a deontological code and they can be held accountable for the consequences of certain decisions —such as in healthcare, for instance—, the conversations about what it means for developers to behave ethically or to nurture their own ethical agency are not occurring [17]. They found in their qualitative study that in many instances, the developers do not think about themselves as moral agents, and similarly, they do not see their profession as intrinsically ethical. However, the implementation of AI tools is steadily permeating all areas of human action, and this is conferring unparalleled authority to the developers who make choices throughout the developmental pipeline. As such, it has become necessary to address developers as important moral agents whose choices can have far-reaching consequences.
Considering this argument, the redistribution of agency mentioned before could impact on the development of moral skills of developers, thus leading —perhaps not yet necessarily to deskilling, as it could be argued that there is no basis to determine existing moral skill— but to an inhibition of moral skilling. This means that, in particular junior developers who have less experience and time in the profession, it would be even less encouraged to nurture more reflective or critical forms of engagement with the technological possibilities and outcomes of the application they are tasked with developing.
In response to this phenomenon, introducing purposeful friction into development workflows could help mitigate the unintended deskilling or skilling inhibition introduced by vibe coding. These practices, defined as ”programmed ineficiencies” [ 18] or ”acts of inconvenience” [19] have been conceptualized as features or steps introduced in systems to keep the human agency in the loop. The objective of such steps is then to reinforce or invite cognitive engagement from developers, at the very least at critical stages of decision-making in which choices made can have an impact on certain fundamental interests of customers (i.e., their rights) or for the long-term sustainability of the product being developed.
There are known examples of positive friction in design, for instance, adding extra confirmation steps when one is in the process of deleting an account. Although this paper will not discuss specific strategies, there are at least three forms of purposeful friction that could help vibe coders to critically engage with their code without entirely losing the advantages it ofers. These are general approaches distinguished by the moment in which they are introduced: before, after and while coding 3.
First, before a coding session, a frictional intervention is to establish which tasks should be carried out using vibe coding and which should be left to traditional programming. For instance, it could be determined that any tasks related to the storage or use of customer data for which security guardrails are necessary should never be programmed using solely vibe code. Second, after the coding session, the coder could prompt the AI-assistant to generate a detailed summary of the session and the code generated that can be employed as documentation or learning tool. This suggestion could address the concern about the technical deskilling of junior developers by harnessing the speed and functionality of AI-assistants to help these developers to acquire experience and learn about the code that was generated. This suggestions, however, assumes that the developers would find desirable to reduce the epistemic asymmetry between themselves and the AI-assistant, i.e., become developers rather than 3It should be noted that it is assumed that vibe coding is being used for product development and not for prototyping or for private projects without reach. In these cases, vibe coding do not raises the concerns discussed throughout this paper just ’coders’. While vibe coding seeks to establish a more conversational form of human-machine interaction, this does not exclude that material engagement with and knowledge of the code still have significant advantages that cannot be substituted by vibe coding.
Third, during the coding process, there could be frictional interventions that introduce deliberate slowness such as having pauses to confirm or accept certain complex or critical modules of code, multiple options being presented to the coder, prompting them to read and actively choose the option that best suits the task. The aim of this kind of friction is to move beyond the question of how something works to the question of why and for whom a developer is building a product in the first place. It is through this engagement that the promise of more democratized access to programming might be better realized, while at the same time addressing critical trade-ofs.
Introducing purposeful friction in development workflows where vibe coding is employed is not without its challenges. It is necessary, for instance, to distinguish between positive and negative friction. Although this paper challenges the assumption that seamlessness in development is always desirable and should be a default practice, this does not preclude the fact that there are types of friction that can negatively impact the developer’s work and the customer’s experience with a product. Similarly, it should be noted that development workflows can vary radically in terms of scope, aims and particular constraints and such factors should be taken into consideration when choosing which types of purposeful friction interventions are implemented. In this sense, the very act of introducing friction is an exercise of reflection about the direction and structure of a particular development workflow.
This paper aims to explore the ethical concerns raised by the emergence and adoption of the practice of vibe coding in software development. It contends that while the alleged advantages ofered by this approach in terms of prototyping speed and lowered entry-barrier for people with little to no programming expertise can be used in certain scenarios, ultimately, vibe coding raises serious ethical questions about the future of software development. These concerns are rooted in the fact that consumers are increasingly using AI-assisted and vibe coding platforms. This creates a series of market incentives and pressures that make vibe coding an attractive —or even necessary—practice for both experienced developers who need to keep up with the fast pace of the development industry as well as junior developers without experience who want to enter it.
The ethical concerns raised are of technical and moral nature. On the one hand, security vulnerabilities and technical debt as a result of poorly implemented and brittle applications could not only jeopardize the privacy and digital integrity of consumers, but also the long-term sustainability of the products themselves, and thus, could impact in the financial health of the companies that developed them. On the other hand, the further obscuring of responsibility channels as well as the re-distribution of epistemic and moral agency, point at more subtle and long-term consequences of practices like vibe coding, that could deskill or inhibit the skilling of professionals in this field not only regarding skills that are required for the technical robustness of the applications but also the moral skills that are necessary for decisions to prioritize goals like securing the rights of people using the products, the environmental costs of developing certain types of software and the long term consequences.
The introduction of purposeful friction can be seem -and perhaps rightly so- as a contradiction to the very essence of what vibe coding stands for. However, the benefits in terms of lower entry barrier and bolstered eficiency of vibe coding do not need to be entirely sacrificed. This paper thus sought to argue that there can and should be a discussion about finding common ground between employing vibe coding and the critical importance of establishing clearer channels of attribution and distribution of responsibility in development settings. After all, given that developers have been granted significant levels of decision-making power in high stake environments, we should at the very least consider that this authority may require stricter standards of professionalization. Ultimately, this paper aims to emphasize the importance of maintaining developer agency in AI-mediated environments from ethical and technical perspectives, and to propose the introduction of purposeful friction as a feasible way to promote forms of cognitive and moral engagement that enable more meaningful configurations of human-machine interaction in which phenomena like deskilling is not an inevitable consequence, but where reskilling and upskilling are also within the horizon of possibility.
The author have not employed any Generative AI tools. [17] T. A. Grifin, B. P. Green, J. V. M. Welie, The ethical agency of ai developers, AI and Ethics 4 (2024) 179–188. doi:10.1007/s43681-022-00256-3. [18] F. Cabitza, A. Campagner, D. Ciucci, A. Seveso, Programmed ineficiencies in dss-supported human decision making, in: P. Editors (Ed.), Proceedings of the 21st International Conference on Human-Computer Interaction, volume 11506 of HCII ’19, Springer-Verlag, Berlin, Heidelberg, 2019, pp. 217–229. doi:10.1007/978-3-030-26773-5_18. [19] R. R. Gosline, Why AI Customer Journeys Need More Friction, Technical Report, Harvard Business Review (online news), 2022.