The Old Way: Weeks of Telephone

Bjorn opened with a familiar scenario. A client asks for a new feature. The product team kicks off research, writes specs, hands those specs to design, and waits. Weeks later, the design team delivers a set of variations. The specs go through another round of review, get handed to engineering, and eventually a first MVP ships. The client sees it and flags thirty things that are not what they wanted.

He used WeGroup as a concrete example: the team once prototyped a customer portal through the traditional process. A big feature like that required extensive spec writing, multiple design rounds, and significant engineering effort before anyone outside the building saw it. The feedback that came back invalidated large portions of the work.

“The stronger and the closer the visual part and even the interactive part is to what the actual product could be, the better and the more qualitative the feedback is.”
- Bjorn Vuylsteker

That insight drives the entire approach. Static mockups produce vague feedback. Interactive prototypes that look and feel like the real product produce specific, actionable feedback. The question is how to build those interactive prototypes without burning weeks of effort.

The Three-Tool Pipeline

Rather than relying on a single all-in-one platform, Bjorn demonstrated a lightweight pipeline of three free tools. Each tool handles one job well, and the handoff between them is straightforward. The pipeline looks like this: generate a visual design, refine it into clean frontend components, then wire in functional AI logic.

Google Stitch: AI-powered prototyping that generates full visual designs from a text prompt. Free to use, quick to iterate, recently gained inline text editing, and exports clean HTML and CSS.
Magic Patterns: imports directly from Figma design systems, producing brand-consistent components. It is a paid tool, but The Product Consortium members get one year of access included in their membership. Requires more upfront setup but delivers better consistency for teams with established design tokens.
Google AI Studio: connects a working LLM (Gemini Flash) to the prototype, turning static screens into interactive experiences with real conversational logic.

lightbulbWhy not an all-in-one tool?

Bjorn was explicit about the distinction: when you build with a full product tool like Lovable, you are building a product, not a prototype. You end up burning through tokens trying to fix deeper layers of logic and infrastructure. A pure frontend prototype is cheaper to change because there is nothing underneath it except the visual layer.

Google Stitch: From Prompt to Design in Seconds

The demo started with Google Stitch. Bjorn described a scenario aloud: one of WeGroup's insurance broker clients wants their end customers to be able to talk with an AI version of themselves. He typed a short prompt into Stitch and within seconds had a complete visual layout for a chat interface, branded and structured.

Stitch recently added the ability to edit text and small elements directly in the generated design, which means quick tweaks no longer require regenerating the entire output. Once the layout looked right, Bjorn exported the result as HTML and CSS. That export is the handoff artifact: a static frontend that looks like a real product but contains no logic.

infoCost and access

Google Stitch is free. There is no paid tier required for the prototyping workflow demonstrated in this talk. The export produces standard HTML and CSS that any downstream tool can consume.

Magic Patterns: Design System Consistency at Scale

Bjorn then showed an alternative path using Magic Patterns. Where Stitch is fastest for one-off explorations, Magic Patterns shines when your team already has a Figma design system. It imports your tokens, colors, and components directly, so every prototype it generates is consistent with your existing brand.

The tradeoff is clear: Magic Patterns is a paid tool that requires more upfront investment in setting up the design system import, but once that is done, every subsequent prototype is automatically on-brand. The Product Consortium members get one year of Magic Patterns access included in their membership. For teams that prototype frequently or need to present to external stakeholders, that consistency pays for itself quickly.

The key advantage Bjorn highlighted is that Magic Patterns produces pure frontend code. Nothing more. If you want to change something, you are only changing the visual layer. There is no backend to break, no database migration to manage, no infrastructure cost. That constraint is actually the feature: it keeps prototypes cheap and disposable.

Google AI Studio: Making Prototypes Talk Back

The final step in the pipeline turned a static design into something interactive. Bjorn opened Google AI Studio, connected Gemini Flash to the exported HTML, and gave the chat interface real conversational ability. The prototype could now respond to typed questions, generate risk analyses, and offer insurance recommendations.

This is the moment the prototype becomes genuinely useful for stakeholder feedback. A static mockup prompts reactions like 'looks nice' or 'change that color.' An interactive prototype that answers questions and displays dynamic content prompts reactions like 'the risk score should weight flood data differently' or 'brokers need to override the AI recommendation here.' The depth of feedback changes entirely.

lightbulbGemini Flash keeps it free

Google AI Studio with Gemini Flash is free at the volumes needed for prototype demos. You are not paying per token for stakeholder walkthroughs. The cost stays at zero until you decide to build the real product.

What Just Happened: Ten Minutes, Zero Euros

Bjorn paused the demo and let the room absorb what had just happened. In roughly ten minutes, starting from a blank screen and a spoken idea, he had a functional prototype with a working chat interface, risk analysis, and insurance recommendations. The total cost was zero. The prototype was shareable via a link.

Before and After

Before: write specs over several days, hand off to design for a multi-week cycle, wait for engineering to build an MVP, then discover in the first client review that the core assumptions were wrong. After: prompt a visual design in Stitch, export the HTML, wire in conversational AI through Google AI Studio, and share a working prototype link within the same meeting. Collect specific, actionable feedback immediately.

The point is not that prototypes replace production engineering. They do not. The point is that prototypes built this way are disposable by design. If the client hates it, you lose ten minutes. If the client loves it, you have a validated direction and concrete feedback to hand to engineering. Either outcome is better than spending weeks on assumptions.

infoThe numbers from the slide

Total cost: zero euros. Time from blank screen to shareable link: approximately 10 minutes. Tools used: all free tiers. Production code written: none.