The internal AI tool that's transforming how Stripe designs products | Owen Williams
Owen Williams, a design manager at Stripe with an engineering background, built an internal prototyping tool called 'Protodash' that lets designers and PMs build high-fidelity, on-brand Stripe prototypes without needing deep coding knowledge. The tool evolved from a set of Cursor rules and a React app into a full browser-based studio ('Protodash Studio') with AI-assisted iteration, design review modes, and dev box hosting. Unexpectedly, PMs became even heavier users than designers, transforming how product work is communicated and reviewed at Stripe.
Summary
Owen Williams, a design manager at Stripe who leads the developer experience space, built an internal AI-powered prototyping tool called 'Protodash' over approximately 18 months. The core problem he was solving was that designers using external tools like V0 kept producing prototypes with mismatched design systems — what he calls 'indigo blur slop' — because those tools didn't know Stripe's internal design system called 'Sail.' His engineering background gave him a pragmatic approach: bundle Stripe's design system components, cursor rules, and an MCP server integration into a starter project that designers could open in Cursor or Claude Code and immediately get 90% of the way to a realistic, on-brand prototype.
The first version was essentially a React router with Stripe's design system components, an MCP server for the Sail design system, and a set of opinionated Cursor rules that taught the LLM how to use the project correctly — including rules to check the design system before writing any code, and explicit prohibitions against hallucinating components or defaulting to Tailwind. Designers could run this locally or, more powerfully, spin up a dev box from a single URL in about two minutes, getting a shareable prototype link without ever running npm.
The second evolution, 'Protodash Studio,' brought the entire experience into the browser — inspired by the goal of building something like V0 but specifically for Stripe. It features an embedded LLM interface for building and iterating on prototypes without needing Cursor open, a visual annotation mode where users can click directly on prototype elements and leave AI-addressable comments, a structured design review mode that replaces Google Doc feedback tables with inline comments and AI-generated summaries, and state/variant controls for quickly switching between business models, data states, languages, or fidelity levels (including a grayscale mode and a low-fi monospace mode).
A surprising outcome was that PMs became the heaviest users of the tool, not designers. Owen initially felt nervous about this but came to see it as transformative: PMs can now explore ideas visually at high fidelity, test prototypes with users earlier, and have more productive conversations with designers about actual work rather than resource allocation. The tool also dramatically changed design-to-engineering handoff — one team working on the Radar fraud detection product built a fully interactive multi-step prototype that an engineer could use as a near-direct source of truth, with all the same components already in place.
Owen built nearly all of this by 'yelling at Claude Code for 18 months,' with his engineering background providing the architectural judgment needed to direct AI effectively. He emphasized that internal tooling has a significant advantage over production software: it doesn't need to be production-grade, which dramatically lowers the bar and accelerates iteration. He is now hiring a design engineer specifically to continue developing Protodash.
Key Insights
- Owen found that PMs became even heavier users of Protodash than designers, which initially made him nervous but ultimately proved transformative — it gave PMs the visual tools to manifest their ideas, made them better communicators with designers, and enabled much earlier user research on prototypes rather than static documents.
- Owen argues that one of the most critical components of making AI prototyping reliable is building extremely opinionated cursor rules — including explicit prohibitions like 'do not use Tailwind unless told to' — because LLMs will otherwise hallucinate entire design systems or default to generic styling without telling the user.
- Owen describes a design-to-engineering handoff transformation where a team on Stripe's Radar product delivered a prototype as a pull request that an engineer could use almost directly, because all components in the prototype were the same ones used in production — something Owen says has never happened in his career as a design manager.
- Owen's architecture insight for Protodash Studio was to wrap the existing React prototyping tool in a browser-based layer so designers never need to open a terminal, run npm, or configure anything — a dev box spins up in about a minute from a URL and the prototype is immediately shareable, eliminating the local machine barrier that previously excluded non-technical designers.
- Owen argues that prototyping complex data dashboards with all their states, filters, zero-data and full-data conditions, and internationalization was 'nearly impossible' in Figma, and that code-based prototyping now lets teams show how real users will actually experience a product — including switching business models, merchant names, or locale in seconds.
Topics
Transcript
[0:00] My dream was I want something that's like vzero but fast. We have all of these tools internally that are really cool. We can connect different data sources together. Why can I not just do this in my browser? Like why do I need cursor? >> You're seeing a lot of designers use it but maybe even more PMS. >> I started seeing PMs use it and got a little nervous. Oh my goodness. PMS designing. It's like what's going to happen? is that how painful is it to prototype a data dashboard with all its interactions, all its filters, all its states, different states, zero data, a bunch of data. It is nearly impossible [0:31] to do that…
Full transcript available for MurmurCast members
Sign Up to AccessMore from How I AI
Claude automates the busy-work so you can spend more quality time with your kids
The speaker discusses how Claude's Co-worker feature helps parents automate tedious online administrative tasks, freeing up time for more meaningful interactions with their children. By handling tasks like returns and help emails, AI removes low-value busywork rather than replacing genuine human experiences.
Use Claude as your personal shopping assistant
A parent describes using Claude as a household management and shopping assistant to find high-quality, naturally-made products from reputable brands. They created a project in Claude with specific brand criteria and used it to organize notes and vet brands. A key benefit highlighted was Claude surfacing that a previously reputable brand had declined in quality after a corporate takeover.
She built a Claude shopping assistant to stop buying cheap junk
Nicole Ruiz demonstrates how she built a Claude project to automate high-quality shopping decisions for her family, using curated brand lists and purchasing criteria to filter out cheap, poorly-made products. She also shows how Claude Computer Use helps her draft return emails by pulling order details directly from her Gmail. The system is designed to reduce the mental overhead of conscious consumption so she can spend more time with her children.
Creating a podcast hype video with Gemini Omni
The host demonstrates using Google Flow, a generative AI creative suite, to create a hype video for the 'How I AI' podcast. Using a fish-eye lens avatar of themselves, they generate a storyboard with cinematic shots including keyboard close-ups, office wide shots, and a humorous chair spin with a digital heads-up display overlay.
She shipped an app to the app store with zero coding knowledge
A person recounts shipping an app to the app store without learning to code, relying primarily on basic computer skills like copying, pasting, and file labeling. She candidly admits she still doesn't understand the underlying software infrastructure, including the deployment platform Railway, even after successfully completing the project.