Source-linked AI summary
Towards Effective Generation of Interactive Visualizations with Vibe Coding: An Empirical Study
Yanshan Zeng, Ruixuan Tu, Zuo Xiang, Lijia Feng, Guozheng Li, Chi Harold Liu
TL;DR
Interactive visualization construction combines programming and design demands, while vibe coding’s capabilities and developer experience in this setting have remained insufficiently examined. This paper studies the issue with 78 participants using construction tasks, questionnaires, interviews, and case analyses. It finds faster prototyping alongside limitations in complex interactions, visual detail, synchronization, and code stability, with developers shifting toward semantic alignment and hybrid collaboration.
Problem
Vibe coding’s capabilities, limitations, and developer experiences in constructing complete interactive visualizations remain insufficiently examined.
Method
The authors conducted an empirical mixed-method study in which 78 participants constructed interactive visualizations and completed questionnaires and semi-structured interviews.
Results
Vibe coding improves prototyping efficiency, especially for static charts and simple interactions, but remains limited for complex interactions, multi-view synchronization, visual details, and code stability.
Takeaways & Limitations
Developers compensate through layered collaboration, task decomposition, semantic alignment, validation loops, and selective manual code refinement.
Takeaways & Limitations
The findings are based on computer science students, non-standardized models and tools, and an initial development process that excludes subsequent optimization.
Abstract
from arXiv · showhide
Constructing interactive visualizations has traditionally required substantial human effort, involving both technical implementation and design decision-making. Recently, vibe coding, a programming paradigm leveraging Large Language Models to generate, interpret, and refactor code from natural language specifications, has emerged as a promising approach to reduce the burden. However, the capabilities and limitations of vibe coding in building interactive visualizations remain unexplored. To address this gap, we conducted a user study with 78 participants that were tasked with constructing interactive visualizations using vibe coding. We further collected users feedback through questionnaires, interviews, and case analyses. Based on this study, we examine (1) the capabilities and (2) user experience of vibe coding in generating interactive visualizations, and (3) the practical human-agent collaboration strategies adopted. Our findings provide the first systematic assessment of vibe coding for interactive visualization construction, revealing both its strengths and limitations, explaining the shift in developer labor and identifying the hybrid collaboration strategies participants adopted. Furthermore, our study offers insights for more intuitive and robust vibe coding practices.
1 INTRODUCTION
Interactive visualization construction combines substantial programming effort with interdependent data, visual, interaction, and layout decisions, making it a challenging setting for vibe coding. This study systematically examines vibe coding’s capabilities, user experience, and collaboration strategies through a 78-participant user study.
- Motivation: Vibe coding lets developers specify goals, constraints, and modifications in natural language while AI agents generate, interpret, and refactor code.It is presented as a way to reduce programming burden and accelerate interactive visualization creation.
- Challenge: Interactive visualizations require coordinated data processing, visual mapping, interaction logic, and layout organization alongside design decisions.These requirements make visualization construction more complex than tasks involving programming alone.
- Study focus: A 78-participant university study examined vibe coding’s effectiveness, user experience, and developer strategies for constructing interactive visualizations.The study used questionnaires, interviews, and case analyses to investigate capabilities, limitations, and practices.
- Findings: Vibe coding accelerates prototyping but struggles with fine-grained interaction, visualization design, and system-level consistency.Participants reported a shift from coding toward semantic alignment, often creating a description-verification bottleneck.
- Findings: Participants commonly combined high-level prompting with selective manual code intervention to address vibe coding’s limitations.This hybrid strategy was identified as a practical collaboration approach.
- Contributions: The study contributes a systematic assessment, a characterization of advantages and pain points, and design implications for future vibe coding techniques.These contributions cover capabilities, user experience, interaction patterns, and support requirements.
2 RELATED WORK
Prior work has studied vibe coding and LLM-based visualization tasks, but has not sufficiently examined developers constructing complete interactive visualization systems. This study addresses that gap through a user-centered evaluation of capabilities, limitations, user experiences, and collaboration strategies.
- Prior work: Existing vibe coding studies have focused on efficiency, validation costs, and interaction methods in general-purpose programming.
- Prior work: Visualization research has applied LLMs to chart generation, multi-view construction, and visualization code synthesis, often as isolated tasks or from end-user perspectives.
- Research gap: Developers’ practical capabilities and user experiences with vibe coding remain insufficiently examined.
- Research gap: Prior work has not systematically investigated collaboration strategies for design decisions and the collaborative processes involved in building complete interactive visualization systems.
- Study contribution: This study addresses the gap through a user-centered evaluation of vibe coding in interactive visualization construction.
3 METHOD
The study combined a one-week visualization construction task with a questionnaire and semi-structured interviews involving 78 computer science participants. Quantitative descriptive statistics and qualitative thematic coding were used to evaluate capabilities, experiences, and collaboration strategies.
- Participants: The study recruited 78 university participants with computer science backgrounds, interactive visualization knowledge, and vibe coding experience.Participants used varied LLM strategies, ranging from near-total reliance to troubleshooting-only assistance.
- Procedure: The study comprised an interactive visualization construction task, a post-task questionnaire, and a semi-structured interview.
- Procedure: During the one-week task, participants developed interactive visualizations with data transformation, visual mapping, exploratory interactions, and responsive interfaces.They could choose their own topics and datasets.
- Procedure: The 20-minute questionnaire captured retrospective assessments of capabilities, user experience, and human-agent collaboration strategies.
- Procedure: The 25-minute interviews investigated issues and best practices beyond the questionnaire.
- Analysis: Quantitative analysis summarized five-point Likert responses with means, standard deviations, medians, and interquartile ranges.Measures covered performance perceptions such as delivery efficiency and code reliability, plus intent alignment and cognitive load.
- Analysis: Qualitative analysis thematically coded open-ended responses and interviews, using initial Gemini coding followed by manual review.The analysis targeted advantages, limitations, and collaborative strategies.
4 RESULT
Vibe coding accelerates prototyping and delivery but performs inconsistently on complex interactions, detailed visualization design, code reliability, and intent alignment. Participants therefore adopted hybrid workflows that combine incremental prompting, constraints, agent assistance, and manual refinement.
- Capabilities: Vibe coding accelerates delivery and prototyping, including rapid setup of basic frameworks and charts, but generated systems still contain bugs.Participants reported advantages in debugging and overall delivery despite residual implementation issues.
- Capabilities: Performance declines as visualization tasks become more complex, personalized, and system-wide, especially when interactions require shared state across views.One-shot generation is strongest for static charts and basic interactions and weaker for complex interactions and multi-view coordination.
- Capabilities: Generated code remains insufficiently reliable for direct delivery, with participants reporting confusing interaction logic, mismatched view synchronization, and improper visual mapping.These issues indicate functional output whose detailed quality still requires manual review.
- User Experience: Developers experience a shift from coding toward semantic alignment, re-prompting, validation, and back-and-forth inspection of the interface after changes.The study reports cognitive friction from unpredictable feedback and switching between agent dialogue and visual previews.
- User Experience: Fine-grained visual and interactive specifications require repeated prompting because natural language does not reliably convey layout, positioning, and complex interaction sequences.Participants identified semantic discrepancies and prompt-construction barriers as prominent sources of refinement effort.
- Collaboration Strategies: Participants commonly decomposed tasks incrementally or supplied sample data, while reference diagrams and lengthy single instructions were less common strategies.Incremental feature addition was reported by 61.5% of participants and sample-data provision by 59.0%.
- Collaboration Strategies: Vibe coding functioned as hybrid collaboration: 59.0% manually modified 10%–30% of code, while only 28.2% expected minor adjustments to suffice.Developers remained responsible for checking and correcting generated details.
- Collaboration Strategies: Participants chose between agent and manual intervention according to efficiency cost, modification granularity, and the cost of understanding the code.Manual edits favored quick local changes; agent guidance favored extensive structural changes or unfamiliar technology stacks.
5 DISCUSSION
The study identifies tool requirements for addressing vibe coding’s difficulties with layout, interaction debugging, and visual version control, while documenting a co-design process in which agent outputs stimulate human design shifts. It also reports scope limitations involving participant background, tool variability, and development stage.
- Vibe coding tools need better support for layout adjustments, interaction debugging, and visual version control.
- An agent-generated bidirectional bar chart shifted one designer from geographic representations toward numerical distributions and a final interactive dumbbell plot.The case concerned state-wise U.S. bipartisan votes over time and compared two selected years.
- Multimodal detail edits should connect direct visual manipulation with underlying code, parameters, and natural-language commands.
- Visual debugging should expose data mappings, state propagation, event sequences, interaction histories, and modification scopes.
- Controlled iterative collaboration should let developers lock components, preview differences, roll back versions, and track design intent.
- 78 students participated, but findings may not generalize beyond computer-science students, standardized tools, or initial visualization development.The study also notes that model, platform, and tool differences may have influenced evaluation results.
6 CONCLUSION
This empirical study evaluates how developers use vibe coding to construct interactive visualizations. It finds improved prototyping efficiency for static charts and simple interactions, alongside persistent limitations that require precise human intervention.
- The study combines quantitative and qualitative methods to evaluate developers’ use of vibe coding for interactive visualization construction.
- Vibe coding significantly improves prototyping efficiency, particularly for static charts and simple interactions, but remains limited for complex interactions, synchronization, visual details, and code stability.
- Vibe coding shifts part of the coding burden toward semantic alignment and validation loops, creating friction during modality switching.
- Developers use layered collaboration, combining task decomposition with selective manual code refinement according to efficiency, complexity, or cognitive load.
- High-quality interactive visualization systems still rely on precise human intervention, motivating better connections between natural-language instructions and visual feedback.