Prototyping with AI as a design partner

AI tools are changing how we design, prototype, and explore product ideas. Not by replacing teams, but by letting individuals move faster, go deeper, and make better decisions earlier. We’ve been testing what this looks like in practice: how designers can build functional prototypes without a developer or product owner in the room.

What follows is not a how-to, but a snapshot of our journey in what worked, what didn’t, and what we have learned to evolve our approach with our product principles. 

How we prototype: start with intent, not wireframes

Rather than begin with high-fidelity mockups, we started with clear prompts. These described the user, the task, and the desired interaction. Sometimes we added screenshots. Sometimes just text. The key shift: design wasn’t something to be handed off. It was something we built directly to be validated with users as quickly as possible.

Three prototypes, three approaches to AI tools

Across three projects, we worked through different AI combinations. Each revealed something new:

1. Multi-user interface prototype
   • Two users in different roles interacting on an app with distinct user flows prototyped in Bolt
   • Combined later in Cursor in a shared code editing environment
   • No backend integration needed initially, but logic was already in place if we chose to deploy
2. Phone plan subscription recommender
   • Started with Cursor, no visual design phase
   • Used GPT to research subscription plans, generate synthetic data, and have Cursor normalize plan data for visual prototyping
   • Iterated UI, and logic purely through text prompting in Cursor to envision and refine the interaction experience with the client before going high fidelity
3. Sales platform with visual depth
   • Initial pass with V0 fell short
   • Rebuilt with Bolt, including custom components and animated avatar
   • Polished in Cursor with detailed prompts and styling

Each one benefited from the same principle: build quickly, shift tools when needed, refine with precision, co-creating with the client.

The emerging pattern for prototyping with AI

We found ourselves using a flexible toolkit, anchored around three roles:

• The raw material: GPT, Claude, Gemini, NotebookLM: Act as a research partner and data modeler
• The hammer: Polymet, Bolt, V0, Lovable: Get a working skeleton fast
• The chisel: Cursor, Windsurf, DesignDoc: Add nuance, behavior, and polish

Prompt clarity mattered more than visuals to start. Screenshots helped, but didn’t carry the context. Structured text almost always produced better results.

Working principles when prototyping with AI

Several principles emerged that now inform our approach:

• Identify constraints and guardrails early (business, technical, brand, timeline)
• Use broad generation tools to quickly scaffold form and layout 
• Switch to refinement tools once structural elements are in place
• Create and reuse prompts like design assets
• Treat AI tools as collaborators, not finishers
• Start small – a single flow or interaction and then expand

What prototyping with AI changes

Prototyping becomes a conversation, not a sequence of tasks. Tools give back ideas. Errors appear sooner. We spend more time refining decisions, not just implementing them. That creates space for us makers to explore, align, adjust, and validate, before committing a team to build.

AI won’t replace product craft today. But it can remove the drag. And when used right, it gives us better results, faster.