The prototype looks finished. The buttons work, the screens alternate, everyone nods their heads. Then development begins and it becomes clear what was missing in the prototype: error state, empty cart, a name that doesn't fit into the field.
AI has accelerated this situation. A prototype can be created in hours, making it easier to approve something that no one has thoroughly checked. The fear of AI workflows is mostly not a fear of technology. It is the fear that no one knows exactly what you will get.
The concern is supported by numbers. The number of job postings on Freelancer.com, where companies are looking for someone to fix AI outputs, increased by 87% from August 2025 to June 2026. The platform's CEO says they are mainly posted by small businesses and entrepreneurs who created the first version through AI and then cannot fix the errors themselves (summary of The Guardian report from 2. 9. 2026).
This can be prevented even before development begins. You just need to know what to ask for from the prototype.
What is a prototype in code and how it works
The classic process ends with handoff: the designer prepares a design in Figma, and the developer must interpret it (in practice, this is called handoff). Everything that the static design does not capture, such as movement or behavior with long text, is supplemented by the developer according to their own estimation. More about where quality is lost during handoff can be found in the article on design engineering.
A prototype in code shortens this step. After Figma, the designer builds a clickable prototype in an AI code editor, such as Cursor. It can connect to the Figma file and read design tokens from it, such as colors, fonts, or spacing (Cursor documentation). The result runs in a browser, so you can try it with movement and states that no one drew in Figma.
Figma remains. It creates the architecture, visual direction, and design system from which the prototype is derived.
Advantages of prototyping in code before development
You can give feedback earlier. Movement, transitions, or error states can be seen before development, not on the finished website. Less translation between roles. When one person leads the concept from design to prototype, part of the fine-tuning after handoff is eliminated. You approve what you can try. Not an image that everyone interprets differently.
How to verify an AI prototype before starting development
Designers themselves state that the biggest obstacle to AI tools is the unreliable quality of output and lack of control over it (AI in Design Report 2026). AI does not approve anything. A human approves, so it is worth asking.
If the agency cannot answer any of these questions, the prototype is not yet complete.
Is prototyping with AI really faster?
Not always, and we prefer to say it right away. A study by METR, in which experienced developers worked on their mature projects, showed that tasks took longer with AI tools, even though they felt they were faster. The authors themselves admit that starting from scratch can significantly speed up the process.
A prototype from scratch is exactly this case. Therefore, we recommend it, but we do not promise acceleration for every task.
For which projects is AI prototyping worthwhile?
The biggest difference is expected where behavior, not just appearance, determines quality: in multi-step forms, configurators, carts, and customer zones. For a simple static website, a prototype in code may not be worthwhile.
AI provides speed today. The designer must guarantee the result.
Do you want to verify the prototype before investing in development? Contact us through the contact form, and we will go through it on your project.
