How Shopify’s Design Studio Uses AI to Accelerate Prototyping
Marvin Schwaibold, who leads the product design studio at Shopify, shared insights into his design process, emphasizing the integration of AI tools and reviewing several viewer-submitted websites.
The Evolving Design Process with AI
Schwaibold notes that while the iterative and sequenced nature of the design process remains unchanged, the tools have significantly shifted. Previously, tools like Figma were central for aesthetic layouts. Now, Figma serves as a "scratch pad" for initial ideas, which quickly transition into functional demos using AI-driven workflows. He primarily works in the terminal with tools like Codex and Cloud Code, enabling faster output and testing on various devices.
The role of the designer has evolved from a "crafter" to an "editor." Designers are now in the driving seat, curating AI-generated outputs and making critical decisions. AI excels at executing complex tasks and writing code, but it lacks the human discernment for what "looks good" or "looks bad." The designer's responsibility is to steer AI in the right direction, breaking down complex concepts into subtasks that AI agents can process incrementally. This shift allows designers to achieve high-quality results much faster, raising the overall ceiling for design output.
Shopify's Product Design Studio and Modernization Efforts
Schwaibold's team at Shopify focuses on redesigning critical user interfaces, from the AI infrastructure layer (Sidekick) that manages the admin panel to the mobile app and admin dashboards. Shopify, being over 20 years old, is undergoing a modernization effort driven by AI.
The design initiative at Shopify has two main components: - Admin Experience Redesign: One part of the team redesigns the entire admin experience, starting with Figma and refining interfaces for the web. - Sidekick Automation: The Sidekick team focuses on automating and manipulating Shopify interfaces through chat. This allows merchants to manage their stores and automate tasks conversationally, reducing the need for direct interface manipulation.
These two seemingly paradoxical approaches work synergistically, creating beautiful new UIs while also leveraging AI for workflow generation, which also enhances the mobile experience.
Molly: A Pre-Shopify Design Studio
Before Shopify, Schwaibold co-ran Molly, a design studio in Brooklyn, with J Tal for two years. Molly worked with major clients like Collins, Apple, and Google, as well as smaller clients, focusing on web design, product design, and content strategy. The studio maintained close client relationships by taking on a limited number of projects at a time.
Molly's work was characterized by organic animations and subtle details, such as ambient lighting effects that matched screen content, inspired by wall-mounted TVs. A key design philosophy at Molly was "primal," a system for structuring website content to encourage exploration rather than demanding attention. This involved progressively disclosing information, ensuring users were never overwhelmed.
An example of this philosophy was a project with Collins, featuring a "bookshelf" section where each book represented a case study. The interactive bounce animation of the books, meticulously crafted by designers and engineers, created a subtle yet impactful moment that conveyed the studio's dedication to craft. This level of detail, while seemingly minor, served as an "Easter egg" for clients, demonstrating the agency's commitment to quality.
Transition to Shopify and New Challenges
The decision to join Shopify was driven by CEO Tobi Lütke's vision for designing the future of the company in a new AI era. Design at Shopify now plays a more prominent role in shaping the company's narrative and enhancing the craft of its tools for merchants. Schwaibold and his team sought a larger impact, applying their design strategies to interfaces used by millions of users.
At Shopify, Schwaibold's studio operates primarily within Sidekick, the AI infrastructure layer. The team works closely with the broader design organization, fostering integration and impact. The process involves extensive experimentation, with designers rapidly building prototypes using AI tools. These prototypes, being highly tangible, quickly gain traction within the company, enabling faster decision-making and buy-in.
AI-Powered Prototyping and Design Exploration
Schwaibold demonstrated how AI tools accelerate the prototyping process. For instance, a redesign of Shopify's admin experience includes "Sidekick suggestions" on the home feed. These suggestions, identified by Shopify, prompt merchants to take actions (e.g., rewriting a product description). Interacting with a suggestion initiates a conversation with Sidekick, which can directly manipulate the admin dashboard and show changes to the user. This aims to bring UI to the user in context, reducing clicks.
Designers at Shopify are actively building these demos and ideas directly into a prototype environment, allowing for rapid testing and iteration. This contrasts with previous methods where extensive hand-crafted designs in Figma would precede engineering involvement.
Schwaibold illustrated his personal workflow with a side project involving a 2x3 grid of videos that could be opened and dragged down to close. He used Codex to build this, but the initial interaction was not "interruptible." To fix this, he sketched out the desired responsive behavior in Figma (e.g., clicking a new box while another is closing) and fed these visual ideas back to Codex. This highlights how Figma acts as a scratchpad for communicating complex motion and interaction concepts to AI, which can interpret visual cues effectively.
He emphasized two key points about AI: 1. Capability: AI software is highly capable of creating anything, but designers must understand the complexity of what they're building. 2. Decomposition: Designers need to break down complex concepts (like a shader) into smaller components, understand the commands, and then reassemble them.
Crafting Lifelike Visuals with AI: The Grass Shader Example
Schwaibold detailed a project where he used AI to transform a video of grass into a controllable, lifelike shader for the web. The process involved: 1. Initial Prompt: Feeding the video to Codex and asking it to create a pixel-perfect shader, matching colors and recreating image parts. The initial result was far from the desired outcome. 2. Iterative Refinement with ChatGPT: Screenshotting the AI's output and the original video, then explaining to ChatGPT why the previous agent failed. ChatGPT helped him understand the need to specify details like light source, shadows, and to break the image into smaller parts for individual rendering. 3. Topographical Mapping: Realizing the AI didn't understand the underlying terrain, he used ChatGPT to render a topographical map of the grass structure. Feeding this back to AI helped it understand depths and heights, enabling more realistic grass rendering. 4. DialKit for Control: Integrating "DialKit" into the demo allowed him to expose and manipulate variables like grass blade length and wind movement. This revealed underlying issues (e.g., a skewed terrain map) and enabled precise tuning of the visual effects.
This project demonstrated that AI empowers designers to execute complex ideas that were previously beyond their individual capabilities, fostering a "full-stack" approach where designers can build and test their creations.
DialKit: Empowering Iterative Design
DialKit is a tool that simplifies the creation and tweaking of variables within a project. It allows designers to easily expose underlying parameters and control them through widgets, enabling rapid iteration and perfect tuning of interactions. This was previously too time-consuming or complex to justify, but AI makes it feasible to build these custom control tools.
Schwaibold showcased DialKit's use in recreating a nuanced carousel animation inspired by Mike Matas. The carousel involved complex physics, like energy ripples and collision effects. DialKit helped him "look under the hood" of the AI-generated code, allowing him to dial in parameters like "pull boost" to achieve the desired organic feel.
Advice for Designers and Founders
Schwaibold's core advice for designers and founders leveraging AI: - Infinitely Patient Collaborator: View AI as an infinitely patient teammate. Don't be afraid to iterate many times to get an animation or design just right. - Tinker and Experiment: Actively experiment with AI tools. Describe what you want, and if the result isn't right, explain why. Provide visual references (websites, screenshots, color palettes) to guide the AI. - Copy and Decompile: For designers, copying existing designs or animations is a great learning tool. AI acts as an "infinitely patient teacher," allowing designers to build and decompile complex interactions to understand their underlying structure without needing to involve engineers. This fosters better management, collaboration, and teamwork. - Set an Unreasonably High Bar: AI tools enable designers to meet incredibly high standards much faster than before.
Website Reviews
Nstack
- Positive: Emulates a desktop environment, allowing users to interact with the product immediately. Feels whimsical and retro.
- Negative: Overloading visual information, making it hard to determine the primary focus or call to action. Too many competing elements (sidebar, chat history, main messaging, multiple CTAs) create a "not dictatish" experience where the user doesn't know what to look at first.
- Feedback: Reduce clutter, prioritize a clear headline and introduction, and offer a simple, quick demo. Focus on one core value proposition rather than trying to do everything at once.
Skill Sync
- Positive: Clear company name and purpose ("skills" and "sync").
- Negative: Small fonts, inconsistent typography, and dense information make it hard to read and understand. Abstracted functionalities behind tabs are likely to be missed by users. The page is too short, hiding important information. AI-generated animations, while technically impressive, don't always add value or clarity.
- Feedback: Use a "three-up component" to highlight the top three features, making them tappable for more information. Extend the page length to progressively disclose information rather than burying it in tabs. Focus on communicating the core value through clear, art-directed illustrations rather than generic AI animations.
Instaplay
- Negative: Lack of clear copy under the title makes it difficult to understand the primary purpose. The visual focus on existing games distracts from the main call to action: creating your own games. Clicking "create" leads directly to a sign-up page without allowing users to experience game creation first. Navigation elements like "my games" and "for you" are unclear and also lead to sign-up. The "explore" button removes the input field, hindering game creation. The fast-forward icon for navigation is confusing.
- Feedback: Clearly state the primary purpose (create your own games) at the top. Allow users to create and play a game before prompting them to sign up. Use interactive carousels with short video reels for game thumbnails to showcase variety. Focus on the value proposition of game creation and sharing.
Mentlify
- Positive: Professional, trustworthy, and clean design. Reductive in the correct ways, showing only necessary information. Clear headline, concise paragraph, prominent "get started" button. Dashboard graphic adds to the vibe without distracting. Strong social proof with a well-designed logo wall of credible companies. Abstracted UI and features into cohesive, art-directed visuals. Well-designed, swipeable carousel with colors mapped to photography. Breathable layout with good spacing between sections.
- Negative: Headlines for feature cards are a bit too small, and the repeated braided animation loses its magic. Important information (e.g., "control who has access") is sometimes buried or overshadowed by visuals. Carousel colors are too close together.
- Feedback: Increase headline size for feature cards. Use the braided animation more sparingly. Differentiate carousel colors. Ensure important copy is prominent and not overshadowed by visuals. Structure the homepage with clear chapters (e.g., products, solutions, resources, pricing) that lead to deeper content.
Nullify
- Negative: Too many font sizes (up to five) in the initial viewport, creating visual clutter. Two changing placeholder elements in the headline are disorienting and don't clearly convey the product's video-centric nature. The headline focuses on the negative (people don't read docs) rather than the positive (videos convert better). The website doesn't immediately showcase its own video capabilities. Carousel navigation is clunky (requires clicking tiny buttons, not tappable cards, no swiping). Videos open full-screen with only an escape button to exit. Scroll jacking makes the experience sluggish and uncomfortable.
- Feedback: Simplify typography to a headline and a single paragraph. Create a concise headline that highlights the video solution. Feature a Nullify-created video prominently on the homepage. Make carousel cards tappable and allow swiping. Implement a more intuitive video viewing experience (e.g., zoom-in, swipe between videos). Avoid scroll jacking; use native scrolling. Prioritize showing what the product does before showing results.
Instinct
- Positive: Very immersive and clear about its focus on space technology. Good use of a serif font for headlines, adding a stylistic touch. Hover animations for "launch, cruise, land" are engaging. Real video content conveys authenticity.
- Negative: The moon in the background is a static photo; could be an interactive 3D WebGL object. Logo and typography in the header are competing. Too much color in partner logos distracts from the main message. Hover animations for "launch, cruise, land" are too slow to trigger. The "land" video doesn't clearly convey its purpose. Important content (technology, company, news, careers) is hidden behind navigation rather than being integrated into the homepage. Inconsistent quality of cover images on sub-pages. Incorrect use of serif font for body copy. Carousel on the technology page is not interactive (no dragging or swiping).
- Feedback: Implement interactive 3D elements for the moon and satellite. Streamline header typography and logo colors. Speed up hover animations for immediate feedback. Refine the "land" video to clearly show the landing moment. Integrate key content sections into the homepage as chapters. Maintain consistent visual quality across all pages. Use sans-serif fonts for body copy. Utilize interactive carousels (like Blossom Carousel) that support swiping and dragging.
Takeaways
- Marvin Schwaibold explains that while the iterative design process stays the same, AI tools have turned Figma into a quick‑scratchpad and moved most implementation work to terminal‑based agents like Codex and Cloud Code.
- He describes the designer’s new role as an “editor” who curates AI‑generated outputs, breaking complex concepts into subtasks so the AI can execute code while the human decides what looks good.
- Shopify’s Product Design Studio tackles modernization through two parallel tracks: a full redesign of the admin experience and the Sidekick AI layer that lets merchants automate tasks via conversational chat.
- By feeding visual sketches from Figma to AI models, Schwaibold can rapidly prototype interactions—such as a responsive video grid or a lifelike grass shader—and iterate instantly, dramatically shortening the time from idea to testable demo.
- He advises designers to treat AI as an infinitely patient collaborator, constantly decompose problems, use tools like DialKit for parameter tweaking, and set exceptionally high standards because AI now makes high‑quality execution faster than ever.
Frequently Asked Questions
What does Marvin Schwaibold mean by the designer becoming an "editor" in the AI era?
Schwaibold means that designers now act less as sole creators and more as curators who guide AI outputs. They break down complex design problems into smaller prompts, evaluate the generated code or visuals, and make aesthetic decisions, ensuring the final product looks good while the AI handles repetitive execution.
How does Shopify’s Sidekick AI layer automate admin tasks for merchants?
Sidekick integrates a conversational chat interface that interprets merchant requests and directly manipulates the Shopify admin dashboard. When a merchant clicks a suggested action, Sidekick generates the necessary code, applies the change in real time, and displays the updated UI, reducing the number of clicks and manual navigation required.
Who is Y Combinator on YouTube?
Y Combinator is a YouTube channel that publishes videos on a range of topics. Browse more summaries from this channel below.
Does this page include the full transcript of the video?
Yes, the full transcript for this video is available on this page. Click 'Show transcript' in the sidebar to read it.
Helpful resources related to this video
If you want to practice or explore the concepts discussed in the video, these commonly used tools may help.
Links may be affiliate links. We only include resources that are genuinely relevant to the topic.