CASE STUDY
AI Coding Assistant
Disclaimer: All designs presented were either designed by myself or in collaboration with designers under my direction. Due to company’s confidentiality, some details of my work can’t be disclosed.
​
Goal:
Enable non-developers to implement custom logic on top of Wix sites - without writing code - by using an AI-powered assistant that generates accurate, context-aware code snippets based on their business needs.
​
My role:
I kicked off the project—from the initial idea to getting decision-makers on board—and led the hands-on UX design. I worked closely with another UX designer, the PM, and the AI and engineering teams to make sure the solution worked for both users and the business.
​
My role also required staying flexible and adapting to ongoing changes in Wix's AI platform capabilities and the evolving quality of the AI model.
​
Segmentation:
The assistant is designed for citizen developers - mainly designers, freelancers, and business owners who use Wix and want to extend its built-in capabilities, but avoid doing so because they’re not professional developers and have little to no coding knowledge.
​
Research:
As part of the research, I gathered insights from the Partners team’s research, collaborated with a BA analyst, and spoke directly with users. Here are the main insights I found:
-
Around half of user feature requests could be solved using custom code.
-
A small percentage of Wix users actually apply custom code, even though many show clear intent.
-
User interviews revealed that people value speed and simplicity but feel overwhelmed by developer tools—especially JavaScript.
-
Many users already try tools like ChatGPT to generate code and paste it into the editor, showing a behavior the assistant can streamline.​
​
Ideation Phase:
The ideation phase included frequent iterations with stakeholders to keep everyone aligned and move fast. I also defined key use cases the model needed to support, worked closely with other teams we depended on, and mapped which components were already supported by Wix’s AI platform and where we needed to contribute.
During the early design phase, we ran usability tests with users and internal dogfooders using low-fidelity wireframes. My role was to create design hypotheses, define briefs with specific tasks to test, and present a summary report with key findings.
Examples of design improvements following the usability tests:
We identified that in addition to allowing users to input free text prompts, it was crucial to proactively suggest relevant actions based on their current state in the editor.
We explored ways to visualize the connection between the assistant’s suggestions and the specific stage of the user’s workflow, helping users understand the context and relevance of each prompt.
One of the ideas was to create a prompt library—a curated set of ready-to-use prompts that users could explore and run to accelerate their work.
Beyond designing the ideal “happy path,” we also considered edge cases—such as situations where the model fails to generate a helpful response—and began ideating how to handle those cases with clarity and guidance.
Final design:
To showcase the final design, I selected a selection of interface screens.
Initial State – The assistant's interface when first opened by the user.
Contextual Link – How the assistant relates to the current stage in the user’s workflow.
Response State – The assistant displays a generated suggestion or action.
View Code – Users can view and copy the code generated by the assistant.
Troubleshooting State – How the assistant handles failed or low-quality responses.
Retrospective:
Making development accessible to non-coders through AI-powered abstraction presented unique and exciting UX challenges—especially in bridging the gap between the AI assistant and the WYSIWYG interface.
​
Throughout the process, the underlying AI platform and model evolved, which required us to stay flexible and continuously adapt the user flow. We went through multiple iterations to ensure the experience remained intuitive and aligned with user needs, despite shifting technical constraints.
Next?
Find me on
