When building a new digital product, the most important goal is usually the same — get to a functional prototype as quickly as possible. That’s the moment when we learn whether an idea works outside of documentation and whether it’s worth investing in further development. It’s not just about speed, but about reducing the risk of heading in the wrong direction early on.
For our clinic kiosk application, we decided to internally compare three different ways of building a prototype. We wanted to see how speed and quality would differ when using AI-only, a traditional workflow with no AI, and a hybrid approach where a developer works hand-in-hand with AI.
This experiment allowed us to look at the same task from three perspectives and evaluate which method makes the most sense for similar projects. Since the client only pays for the solution that goes to production, we were free to explore all approaches without increasing their budget.
I was part of the team testing the developer + AI workflow. And while I had certain expectations, some results were far more surprising than I anticipated.
The brief: a self-check-in kiosk for clinics
The project had a clear goal: build a touch-friendly kiosk application for clinics that allows patients to quickly and intuitively confirm their arrival — without waiting at the reception desk. Key requirements included:
- large touch-optimized interface elements,
- multi-language support,
- automatic return to the home screen after inactivity,
- integration readiness for secure API communication and ID-card scanning.
Three development paths
AI-only: using several tools with minimal human input (Lovable, GitHub Copilot, Figma plugins such as Anima and the MCP server, GitHub Spark).
No AI: a traditional development process delivered by our partner company.
Developer + AI: a hybrid workflow where AI accelerates the work while the developer ensures quality and correctness.
This is where I was involved. My task was to implement the front end with AI support. Using precise prompts, I generated components based on the Figma design and connected them to the API. My colleague handled the back end, also supported by AI when implementing application logic and integrations. At the end, we merged everything into a functional prototype ready for testing.
How front-end development looked in the developer × AI workflow
When I joined the project, I already had a complete Figma design. Thanks to Figma’s MCP server, I was able to give AI direct access to the entire design — including typography, colors, spacing, and visual styles.
The AI then generated components exactly according to the technical requirements I defined — in this case, React components using Tailwind. Because the design was relatively straightforward, the output was very accurate and required only minimal adjustments.
This meant I didn’t have to spend time on repetitive work and could focus more on testing and fine-tuning details to make the user experience as smooth and intuitive as possible.
I was also surprised by how quickly I managed to implement Trust1 Connector — a service for reading ID cards. All I had to do was describe the service to the AI, send a link to its documentation, and outline how it should work. The AI handled the rest, and the result was functional much sooner than expected.
Refactoring and code unification were significantly faster as well — whenever broader changes were needed, I didn’t have to open file after file. The AI identified all relevant occurrences, understood the relationships between components, and modified the code consistently across the project.
In roughly 24 hours, I had a fully functional front end ready for real-world deployment.
How the other teams approached the task
AI-only: speed without control
The AI-only team tested multiple tools. The most successful outputs came from:
- Lovable: created the basic structure including multi-language support and the inactivity timeout, but any larger code change caused unpredictable shifts in layout and behavior.
- GitHub Copilot: generated the project skeleton, but the output was inconsistent and overly verbose.
- Anima plugin: exported screens from Figma, but lacked responsiveness — a critical feature for kiosk devices.
Each tool produced a clickable prototype, but all of them had significant drawbacks. None were suitable as a foundation for long-term development, so they served mainly as a visual reference for discussion.
Generated home screens:
Generated patient identification screens:
No AI: stability at the cost of time
The partner team followed a traditional development approach. Their output was stable, well-structured, and easy to test. However, development took significantly longer and lacked details that make user navigation faster and clearer — such as auto-filling common data or stronger visual guidance.
Even with careful attention to the Figma design, small deviations in spacing or text alignment appeared — details that AI handled with surprising precision.
Conclusion: Where AI shines — and where it doesn’t
After reviewing all three approaches together with the client, we agreed that the developer + AI workflow offered the best balance of speed and quality. AI accelerated repetitive tasks, while the developer ensured context, logic, and user experience were correct.
In short:
- AI alone cannot deliver a high-quality application — but it can significantly speed up repetitive work.
- It is most useful during early implementation, refactoring, and testing rare scenarios.
- Without a developer, AI lacks context and often produces inefficient solutions. Together, they work extremely well.
Today, the developer + AI workflow is — in our experience — the most effective approach, especially when building an MVP (a Minimum Viable Product that quickly validates whether an idea is worth scaling).
The final kiosk application has passed internal testing and is now ready for deployment in real clinics. Its launch is planned for the upcoming months.
If you’re considering AI in your development process, reach out. We’ll help you apply it meaningfully and effectively.
Frequently asked questions about AI in software development
Can AI build a complete app without a developer?
No. AI can quickly prepare a demo or basic prototype, but it cannot ensure stability, precise logic, security, or full design consistency. A developer is essential.
Why can’t AI prototypes be used as a production foundation?
They tend to be unstable, design-inconsistent, and sensitive to changes. They’re excellent for early discussions — but not as the base for long-term development.
Where is AI most helpful in development?
Mainly at the start — creating structure, generating components, refactoring, and speeding up repetitive tasks.
Why is traditional development slower?
It provides the highest stability and control, but building the first usable prototype takes significantly more time compared to workflows that leverage AI.
What makes the developer + AI workflow different?
It combines rapid generation with expert control over logic, UX, and quality — resulting in a fast, reliable prototype ready for real development.
Is AI suitable for high-security or complex projects?
Yes — but only under the supervision of an experienced developer. AI speeds up routine tasks, but critical parts like security, integrations, or data handling must remain human-controlled.
Can AI replace a designer or reproduce Figma layouts perfectly?
Not fully. It can approximate the design, but lacks precision and cannot guarantee consistent styling or typography. Manual review is still required.
When does it make sense to start a project “AI-first”?
For MVPs, internal tools, or early concept validation. In those cases, AI brings fast, low-cost results.
How much time can AI save during development?
In early phases, often days or weeks. For complex products, the impact is smaller but still valuable in repetitive or large-scale refactoring tasks.
Does AI reduce the overall project cost?
It can reduce costs during prototyping and validation. In production, it mainly optimizes developer workload rather than lowering the total price.
Is AI-generated code secure?
Not always. AI cannot reliably identify security risks or choose the best practices. Critical sections — such as authentication or handling sensitive data — always need a developer’s review.
Will AI eventually create complete applications on its own?
Possibly for simple apps. But for complex, business-critical products, developers will remain essential for architecture, integrations, and accountability.
