Intercom Figma MCP and Claude prototyping — Sneak Peek talk

What the video covers

Domingo Widen, Staff Product Designer at Intercom, demonstrates the workflow his team built for AI-powered prototyping using the company’s own design system. Instead of using Figma Make or standalone AI tools, Intercom’s designers prototype directly in code using Figma Code Connect, Claude skills, and Cursor, with the design system acting as the foundation. The interview also covers how Widen trains other designers on AI tools through an internal prototyping hub.

Who it’s for

Product designers at companies with mature design systems who want to move prototyping from static Figma screens into working code, using AI to bridge the knowledge gap. The episode is especially useful for design leads responsible for training teams on new AI workflows.

Key takeaways

1. Code Connect makes design system components accessible to AI. Intercom ensured every Figma component is connected to its code counterpart under the hood. When Claude generates a prototype, it uses the actual production components rather than generic HTML, resulting in prototypes that look and behave like the real product.

2. Claude skills encode team-specific workflows. Widen created custom Claude skills that package specific tasks — like generating a dialog or exporting a prototype — into repeatable commands. These skills ensure consistency across the team and reduce the learning curve for designers new to AI-assisted code generation.

3. Designers prototype in code, not Figma Make. The team made a deliberate choice to skip Figma Make for prototyping and instead use code environments that engineers also use. This means prototypes can be directly handed to engineers without translation, and designers build familiarity with the codebase.

4. An internal prototyping hub scales knowledge. Widen built a living documentation hub with video tutorials, step-by-step guides, and examples for every AI tool the team uses. This hub became the primary way new team members learn the workflow, rather than relying on one-on-one training.

5. AI multitasking changes the prototyping rhythm. Widen describes running multiple Claude instances in parallel — one generating a component while he reviews another. This concurrent workflow dramatically compresses the time from concept to testable prototype.

Worth watching if…

You want a detailed example of how a design team at a well-known product company (Intercom) has structured their AI prototyping workflow around a design system, including the organizational side of training and documentation.