Is AI a genuine accelerator for product development, or just a novelty? To find out, I conducted a 10-day stress test using Figma Make as part of the Contra competition, with the goal of creating one functional prototype per day. The experiment yielded nine distinct projects and, more importantly, a clear framework for leveraging AI in a professional design workflow.
This isn't just a project showcase; it's a playbook on the practical application, strategic value, and current limitations of AI-powered prototyping.
The AI Prompting Playbook: Vague for Vision, Specific for Support
The most critical factor for success is understanding how to prompt the AI at different stages of the design process. My findings show a clear duality.
1. Use Vague Prompts for Ideation & Core Logic
AI excels when given high-level, conceptual goals based on established patterns. For instance, the prompt "Create a fractal simulation that works on desktop and mobile" was incredibly effective. The AI leveraged its knowledge of fractal geometry and responsive design to generate a robust starting point in minutes. This is ideal for quickly validating the core of an idea.
2. Use Specific Prompts for Code Refinement
Conversely, AI struggles with granular, context-dependent visual tweaks. A prompt like "Increase the padding on the main container" would often fail or produce unintended side effects. The solution was to use the integrated code view. I would make the specific CSS change myself and then use a prompt like, "Refactor this component to be more efficient" or "Explain this function to me." Here, the AI acts as a coding assistant, not a visual designer.
A Strategic Pivot: Identifying Where AI Delivers Real ROI
I began the challenge by attempting to build games. This quickly revealed the tool's limitations. Projects requiring complex game logic, heavy canvas rendering, and bespoke graphics were slow, buggy, and difficult to control.
This led to a strategic pivot towards simulations and data-driven tools. This is where the ROI of Figma Make became clear.
AI Strengths ✅
- Mathematical & Algorithmic Concepts: It flawlessly generated complex visualisations like Boids flocking simulations, vector fields, and fractals.
- Boilerplate UI & Structure: It rapidly scaffolds standard application layouts, forms, and control panels.
- Rapid Idea Validation: It can create a functional proof-of-concept for a tool, like a multi-timezone scheduler, in under an hour.
AI Weaknesses ❌
- Complex, State-Dependent Logic: Not suitable for intricate game mechanics that require a dedicated engine.
- Bespoke Visuals & Pixel-Perfection: Struggles to integrate custom graphics or adhere to a strict design system without direct code manipulation.
- Performance-Intensive Tasks: High-frequency rendering or complex physics simulations quickly hit a performance ceiling.
The lesson is to use AI to automate the known quantities (like algorithms and layouts) so you can focus your human creativity on the novel aspects of the project.
Full Workflow Integration: From AI to Deployment
A prototype is useless if it exists in a vacuum. My workflow demonstrated how to integrate AI into a realistic development pipeline:
- Ideation & Scaffolding (Figma Make): Generate the core prototype.
- Code Extraction: Download the complete project files (HTML, CSS, JS).
- Refactoring & Automation (Claude/Cursor): Use a second AI tool like Claude within Cursor to clean up the code. I even had it write Python scripts to automate the setup process for all nine project folders, saving hours of manual work.
- Deployment (GitHub Pages): Host the final, cleaned-up projects in a public portfolio.
This process shows that Figma Make is best viewed as a powerful starting point, not the final step.
Key Takeaways for Product & Design Teams
AI as an Accelerator, Not an Author
Use AI to build the first 70% of a concept rapidly. The final 30% of polish, nuance, and brand alignment still requires skilled human designers and developers.
Your Domain Knowledge is the Limiter
My attempts to simulate beach erosion and weather patterns failed. Why? Because I didn't have the expert knowledge to write effective prompts. The quality of the output is directly proportional to the quality and specificity of the user's domain expertise.
The ROI is in Speed of Exploration
The true value of this tool is the ability to explore and discard ideas at a fraction of the traditional cost. We built and tested nine concepts in the time it might normally take to fully scope out one.
Looking Forward
AI-powered prototyping tools like Figma Make represent a significant shift in how we approach the early stages of product development. They're not replacing designers and developers—they're amplifying our ability to explore, validate, and iterate on ideas at unprecedented speed.
The key is understanding where AI excels (rapid scaffolding, algorithmic solutions, boilerplate generation) and where human expertise remains irreplaceable (domain knowledge, creative direction, and strategic decision-making).
Thank you for reading!
Want to work with me? Feel free to contact me!
...or just say hello on my social media.