Xiulung Choy - Head of Design at Graphite

In April, Mitul Shah (Design Engineer at Vercel), Cameron Collis (designer at Cursor), and Jamey Gannon walked a packed room through their actual workflows. Mitul demoed his design-to-production process, a four-stage loop he uses to ship features, illustrated through his redesign of Vercel Workflows and the Slackbot he built on top of it. Cameron took the room through a recent project at Cursor end to end, from low-fidelity Figma to the custom prototype playground his team uses to the production monorepo. Jamey, one of the most respected AI-forward creative directors working today, closed with a foundation anyone design-minded can use to get better outputs from generative tools. Mitul Shah, Design Engineer at Vercel What we covered * Mitul’s design-to-production process: a four-stage loop he runs every project through (Context, Design, Code, Craft). * His redesign of Vercel Workflows: the new runs table, the new trace viewer, and what changed between the two-day MVP and today. * Yogurt, the Slackbot Mitul built on top of Workflows, and how using it taught him what to fix in Workflows. * Where agents stop being good enough, and where deep CSS and craft still beat anything an agent can produce. * Q&A: annotations on static images as agent instructions, trusting an agent you only see through Slack, and advice for designers moving into design engineering. Key takeaways * Coding has compressed to the point where it is no longer the bottleneck. The time has shifted to context-gathering and to the last ten percent of craft. * Dogfood the product you are designing. Mitul built Yogurt on top of Workflows, and using Yogurt taught him what to redesign in Workflows. The recursion is the feedback loop. * Craft is specific, not abstract. “How many of you guys have tried to build a sticky header on an HTML table? It’s really hard. And I promise you the agents can’t do it yet, and that’s where I’ll step in with my deep CSS knowledge.” * Annotations on a static image are a useful agent-feedback mechanism. A label on the canvas plus a skill file gets the agent to do the right thing most of the time. * “I am more bullish on design than I am on programming.” Mitul built Yogurt without looking at the code. Cameron Collis, Designer at Cursor What we covered * A recent Cursor project: the self-driving PR concept inside the Review tab, where checks, BugBot, and human reviews feed into a queue an agent automatically fixes. * Why Cameron is still heavily using Figma for early-stage ideation, in black and white. * The custom prototype playground he built after the Graphite acquisition, and how the broader Cursor design team uses it. * The handoff between playground and the production monorepo, and what changes when the monorepo is heavy. * Q&A: when to ship straight to production, the translation problem from playground to monorepo, and how to stand up a similar tool inside another org. Key takeaways * Figma still wins for early-stage exploration. The agent has not yet replaced what Figma does at the start of a project. Cameron: “I find there to be a lot of friction and I cannot stay in a flow state. I’m constantly spending too much time and energy steering the agent and not enough time really thinking through an idea.” * The prototype playground is “about an 85 to 90 percent match” of the Cursor web app. Believability is what earns engagement from the team that a static Figma file cannot. * Vision work and design direction go to the playground. “I know what I want, get in, build it, get out” goes to the monorepo. * On building internal tools without org buy-in: “Just build it if you want to build it. Probably don’t get buy-in. But don’t make your other projects fall apart.” Jamey Gannon What we covered * A foundation any design-minded person can use to get better outputs from generative AI. * Four principles, each illustrated through a single fintech art-direction brief (3D illustrations, social mockups, web). * Jamey’s Midjourney setup: profile codes,

Dylan Babbs, Co-Founder and CTO of Profound, represents a rare breed in tech: the design engineer turned founder who went from redesigning Uber’s navigation system to co-founding New York’s newest AI marketing unicorn in just 18 months. His journey from maps and data visualization to helping Fortune 500 companies stay visible in an AI-driven world reveals what happens when deep craft meets entrepreneurial ambition. At a glance… Dylan describes his philosophy of not compromising on quality even when scaling rapidly, believing that design serves as a company’s “front door” and first impression that can kickstart the entire growth flywheel. He advocates for hiring younger, passionate employees over experience alone, noting that Profound’s average age is around 26-27 and that all current department heads joined as individual contributors before growing into leadership roles. The company operated with a mandatory six-day work week for the first 14 months, trading time for the ability to maintain hiring standards and product quality. Dylan explains that AI represents a new customer that every brand must serve, fundamentally changing how companies approach digital visibility from search optimization to “Answer Engine Optimization.” He emphasizes that successful AI product design requires specialization in at least two of three areas: visual design craft, product design for business impact, and design engineering implementation. The agent builder platform demonstrates Profound’s evolution from analytics to a full “workbench” where marketers can create, orchestrate, and automate AI-powered workflows at scale. Dylan argues that inflection points like the rise of LLMs create opportunities for new companies to succeed where existing defenses no longer apply, advising founders to be mercenaries rather than missionaries when choosing what to build. From designer to founder: the path beyond traditional design roles Dylan’s transition from design engineer at Uber to co-founder reflects a broader evolution in how designers can leverage their skills. “I never really liked having to just not make things real,” he explains, describing how his informatics background combining design and engineering led him to always want to bring designs to life with code and real data. His four years at Uber, redesigning navigation systems and programmatically building 256 design system components, taught him that design principles must serve business metrics. The founding journey involved two years of failed side projects before landing on Profound, driven by what he calls the desire to “advance your own skills” and avoid looking back with regret about missing the AI inflection point. Design as company DNA and competitive advantage Profound’s investment in design from day one stems from both personal passion and strategic necessity. “If you’re a potential customer looking at Profound, the first thing you’re going to see is the website. It’s the front door,” Dylan explains. In the anxiety-filled AI space, premium design experiences build trust with enterprise customers. The investment creates a hiring advantage since top designers don’t want to work for founders who don’t take the craft seriously. Dylan frames this as exploiting a unique advantage: “We’re all designers here. If you want to actually invest in your own design capabilities, it’s going to be very hard to attract that talent if it’s not already in your DNA.” AI as customer: designing for artificial intelligence alongside humans Dylan’s most provocative thesis is that “every company on the planet just got a new customer: AI.” This reframes design from human-centered to intelligence-centered, though he argues the principles remain similar since AI models are trained on human preferences. “You’re not designing directly for the human, but you’re designing for something that represents the human,” he notes. This shift manifests in Profound’s core product: helping brands optimize how they appear in AI-gener

Jenny Wen, Design Lead at Anthropic, joined Double Diamond for a conversation about designing Claude, one of the most widely-used AI assistants in the world. From her early work on Dropbox Paper to bringing FigJam from concept to launch at Figma, Jenny has spent her career building collaboration tools that feel both powerful and deeply human. Today she’s applying that experience to designing interfaces for intelligence itself, leading design for Claude Cowork and helping shape how millions of people interact with AI. At a glance… Jenny describes designing AI products as fundamentally different from traditional software because you can’t map out all possible user flows when dealing with large language models. Where traditional products have deterministic states, AI products offer core primitives with essentially endless interaction possibilities. Her team discovered that knowledge workers are rapidly becoming more comfortable with AI-specific concepts like memory, context, and agents, leading to decisions about using technical terminology rather than abstracting it away. She demonstrates how she uses Claude extensively in her own design process, calling it especially effective at “garbage in, treasure out” analysis of large datasets from user interviews to social media feedback. The rapid advancement of AI coding capabilities has fundamentally changed how product teams operate, with single engineers now capable of shipping entire features end-to-end. As models become more capable at longer-running tasks, Anthropic has started hiding more technical details because users increasingly trust Claude enough not to need visibility into every step. Jenny emphasizes that while shipping code has become accessible to everyone through AI assistance, craft and the ability to execute well remain the most valuable designer skills. The fundamental difference between designing traditional software and AI products Jenny explains that designing AI products requires a completely different mental model than traditional software. “When you are designing now with these LLMs, you basically can’t map out all those flows. You basically just have these core primitives and these specific states that you want the model to get to, but the possibilities in which the user can actually interact with the model are just basically endless.” She compares this to her work at Figma, where despite having millions of possible design outcomes, the team still designed around specific use cases while remaining open to discovering new ones through user behavior. The challenge becomes designing the variables rather than controlling specific outcomes. Building Claude Cowork through rapid iteration and dogfooding The development of Claude Cowork reveals Anthropic’s approach to AI product development through continuous iteration and internal testing. Jenny describes their policy of putting experimental features in front of employees immediately: “Anything goes when it comes to dogfooding internally. We have the most janky stuff just running internally as long as it doesn’t disrupt people’s workflows.” The team experimented with multiple technical approaches and agent architectures before recognizing product-market fit when users started applying Claude Code to non-coding tasks. What became the “10-day ship” was actually the culmination of months of prototyping and internal feedback, with the final decision to ship happening when they saw clear user demand. Knowledge workers are rapidly adopting AI-native concepts Jenny has observed a significant shift in her target users’ AI literacy over just the past year. “Even like a year ago I was uncertain about whether we should reveal words like memory and context and agents or should we abstract them in a way where they’re sort of fluffy. But now I think people know those words.” She explains that knowledge workers at AI-adopting organizations have become much more comfortable with technical AI concepts than expected. Rather than sim

Building the platform layer for AI-native apps People don’t open apps the way they used to. Increasingly, they tell an AI what they need and expect the right tool to just appear. John Allen, Co-Founder and CEO of Layo, is building the platform that makes that possible - turning existing products into AI-native experiences that live inside ChatGPT and other AI systems. From OpenAI’s app announcement in October to having paying enterprise customers in early 2026, John has moved with urgency matched to his conviction: the next operating system is a conversation. We sat down to talk about how he’s building the infrastructure underneath it. At a glance John sees ChatGPT apps as the beginning of a fundamental platform shift, comparable to the early iPhone app store but happening at internet speed. His thesis centers on the shift from screen-driven workflows to intent-driven conversations, where users express what they want naturally rather than navigate complex interfaces. Traditional web and mobile development tools can’t handle non-deterministic outputs or intent-based interaction patterns, creating a new category of infrastructure needs. Layo provides the visual builders, analytics, and configuration tools specifically designed for AI-native apps, helping enterprises ship into the ChatGPT ecosystem in days rather than months. The deeper transformation John envisions is software that adapts in real time to individual user preferences, enabled by AI platforms that maintain context across all applications. Enterprise customers consistently ask the same two questions: what should we build and when, with his answer being “the sooner the better” due to dramatically lower development costs and first-mover learning advantages. The analytics revolution enables something unprecedented in software history: full insight into user intent, measuring why people want to use your product rather than just what buttons they clicked. This leads toward a future of millions of interface variants rather than today’s A/B testing approach, where every user gets software built for exactly how they want to use it. Defining AI-native: intelligence and action as the foundation John draws a clear distinction between companies that “bolted on AI into their existing software” and those building AI-native products. “I think the real definition of AI-native are apps that actually have true intelligence as the baseline foundation. Intelligence and action, right?” He compares it to the data-native era of the 2010s, where companies competed on proprietary data. Now, data is getting commoditized and “all these apps are starting to talk to each other.” The value shifts to becoming “the system of action where people can go and do things” with global intelligence as the foundation rather than siloed data. The operating system shift: from apps you open to systems that respond John frames the current moment as comparable to previous platform transitions from desktop to web to mobile. The fundamental change is that “instead of just having the foundation of like we are the system of record, right now you’re like we’re the system of action, we’re embedded into experiences, meeting customers where they already are.” ChatGPT represents the first true AI operating system where users come to express intent. “People use software to go and do things or to get information” and AI interfaces can meet users at that moment of intent without forcing them to learn new interaction patterns for every application they use. Why the “mobile moment” parallel matters: winning through learning, not timing Drawing explicit parallels to the iPhone app store launch, John believes “the first thousand apps in the app store, they’re going to win the biggest. Not because they were the first thousand, but because they were the first to learn.” The ChatGPT app store launched December 17th, making this moment “like the cutting edge, like really day one.” The advantage comes from accumulating learning cyc

At a glance… When Carl Rivera became Shopify’s first Chief Design Officer in eight years, he inherited a team of 170+ designers across one of the world’s largest commerce platforms. But rather than optimize the existing system, Rivera chose transformation. His approach combines philosophical clarity about design’s role with hands-on building, from shipping code himself to creating internal tools that replace industry standards like Figma presentations. What makes Rivera unique is how he operates simultaneously as strategic leader and individual contributor. During our evening at the Flatiron studio, he live-demoed internal tools his team built while explaining the organizational philosophy behind eliminating job titles, restructuring teams, and requiring all designers to ship code. This is leadership through example and systematic change. Rivera discovered massive latent talent at Shopify that was operating at 70% capacity due to permission culture, which he unlocked by simply stating that design matters and giving people permission to do their best work. The company abandoned universal design processes in favor of material-driven approaches where each project gets the methodology it needs rather than forcing everything through the same workflow. Over 50% of Shopify’s designers now actively ship code to production after completing mandatory engineer onboarding, with Rivera viewing technical fear as the primary barrier holding back the design discipline. Teams have been restructured around “rovers” who move fluidly between projects and agency-style deployments that can pivot quickly to emerging opportunities rather than permanent assignments. Rivera believes taste, aesthetics, and strong points of view become the key hiring differentiators as AI democratizes baseline design skills, favoring candidates who provoke strong reactions over safe consensus picks. The apprenticeship program reflects his commitment to developing junior talent while the industry faces a K-shaped distribution where seniors accelerate with AI but juniors struggle to find opportunities. Internal tools like Artifact demonstrate his philosophy of building what you need rather than accepting limitations, combining project management with presentation capabilities that seamlessly integrate Figma designs, live prototypes, and traditional slides. Permission culture: unlocking latent talent through leadership clarity Rivera’s first major insight as CDO was recognizing that Shopify had tremendous design talent operating below capacity. “There was a ton of latent talent in the company in design specifically. There were a lot of people that were really outstanding but were producing at about 70% of their capabilities,” he observed. The root cause was permission culture where designers felt constrained from doing their best work. Rivera’s solution was direct leadership: “Just saying that design actually really matters and we’re going to become the best at it. Just saying that, my experience was that a lot of people just immediately became better designers because they felt that someone said that they were allowed to.” This experience shaped his broader philosophy about claiming leadership before having the title. The transformation demonstrates how clear communication about priorities can unlock existing talent without hiring. Material-driven process: why Shopify abandoned design frameworks Rather than implementing universal design processes, Rivera took the opposite approach based on his belief that process should follow material. “Companies generally try to come up with the process because it feels like management, and then they take all of their material and squeeze it into that process,” he explained. At Shopify, working on Shopify Payments (processing billions of dollars) requires completely different methodology than launching experimental merchant tools. Rivera’s approach demands active management to understand each project’s unique needs and adapt accordingly.