Katarina Batina - Making big bets with design

2026-05-12 - 55 min - source - Read full transcript

Vid (host)Katarina Batina

Key insights

Shop's biggest product bets (the storytelling feed, the shop store, the undercart drawer) succeeded by deliberately rejecting standard commerce best practices like information density and micro-optimization.

Katarina describes internal pushback from product managers and data scientists when early feed prototypes 'thwarted' conventional density-and-conversion best practices. The team chose a 'window shopping' model instead, one brand story per viewport, betting that a strong enough recommendation algorithm could support lower density. Amazon and Target subsequently moved their own discovery feeds toward the same storytelling-card format, which she reads as validation.

risk-taking-and-big-bets

Taking a big bet on an established, high-stakes surface requires explicitly resetting the team's tolerance for risk and accepting the new version may underperform the incumbent at first.

She cites Jess Ericson's redesign of the Shopify signup page, 'the lifeline of growth,' where the team had to agree upfront that a more brand-forward design might not perform as well as the hardened incumbent. It shipped net positive, but the key move was committing beforehand to iterate on the new version as much as they had iterated on the old one, rather than expecting a first-shot win.

risk-taking-and-big-bets

Designers at Shop now prototype directly against production data using AI coding harnesses (Claude Code), building whole conceptual product directions rather than single end-to-end features.

She describes designer Luke Deon's prototype for a new 'shop store' concept: not one feature built end to end, but a whole configurable world of merchant storefronts pulling from real online-store data. This was made safe by developer-experience tooling and hosted internally on a tool called Quick for instant link sharing, avoiding prototypes that only live locally.

ai-prototyping-workflow

Shop's designers are shifting from React Native to native Swift and Kotlin because those languages are more legible to AI coding agents and to designers themselves, and this technical direction was pushed by design, not engineering.

Designers hit a wall prototyping in React Native due to layers of abstraction and an illegible codebase, so some built a parallel version of the app in Swift, which unlocked an explosion of designer-driven prototyping. That success is now driving Shopify's CTO and Toby to reconsider the platform's committed tech stack, an example of a technical architecture decision being driven by which language is easiest for non-engineers and agents to work in.

designers-writing-code

Designers should ignore feasibility and fully render their idea in whatever medium makes it most believable, without worrying about whether the prototype can ship to production.

Her advice to designers who get stuck debating engineers or fighting the codebase: focus on desirability, the thing design is uniquely responsible for, and use the prototype as a artifact to bring to engineering rather than a production deliverable. She warns against designers only 'designing to their local skill set' and constraining big ideas because of what they personally know how to build.

designers-writing-code

The counterargument to 'ignore feasibility by coding it' is that a rough sketch of a wild idea is actually a purer form of ignoring feasibility than a working code prototype, since code implicitly boxes the idea into what's buildable.

The host pushes back that embracing code as a medium may cap the craziest ideas because they become harder to realize in code; Katarina agrees with this tension and says the risk she actively coaches against is junior designers arriving at critique already listing engineering or data constraints instead of presenting their full vision.

risk-taking-and-big-bets

A multi-artifact workspace that collects prototypes from any tool (Figma, Origami, Swift, web, video capture) into one shared canvas prevents designers from getting stuck trying to force an idea into a single tool that isn't suited for it.

She cites the undercart drawer interaction, which was made in Origami rather than code because it needed precise motion, as an example of why Shopify built 'Artifact,' a tool for centralizing heterogeneous prototype formats in one place so teams aren't stuck in a 'death loop' trying to build an idea in the wrong medium.

ai-prototyping-workflow

As AI-accelerated prototyping produces far more ideas than the team can ship, editing and disciplined focus on a small set of core primitives has become one of the most important parts of the design job.

She says Shop is now generating more ideas than ever, a reversal from when the team bet everything on one or two big yearly moonshots. The discipline now is deciding which of the app's primitives (package tracking, the feed, agentic search, the Shop Pay identity layer) actually drive the most valuable sessions, and resisting adding delightful-in-isolation features that together overwhelm the experience.

craft-and-editing

Annual moonshot planning cycles are being replaced by continuous, week-to-week idea validation, because AI prototyping makes it fast enough to test many directions and 'chase heat' on whichever one gets people excited.

Where Shop used to sit down every December to define a single yearly product vision that shaped the following quarter's roadmap, that cadence now happens in micro-doses. A designer builds a prototype, tells a coherent story in a quick async video, and if people's reaction shows real excitement ('hair on fire'), the team pulls in engineers and chases that idea instead of following a pre-set structure.

ai-prototyping-workflow

As AI compresses the time designers spend on manual execution, articulate critique and design vocabulary (hierarchy, balance, symmetry) become the primary lever for improving work, because feedback to agents is now given in words.

Katarina argues the design profession's material language for building is 'becoming words': designers must be able to say precisely why something isn't right instead of gesturing at a vague feeling, because that same fidelity of language is what lets them direct an AI agent effectively. She notes many self-taught software designers lack this classical design vocabulary and can now use AI itself to learn and articulate these concepts.

craft-and-editing

Great design leaders in the AI era need both a deeply intense relationship with the business and a deep relationship with craft, and staying close to the craft (not managing from a distance) is what lets designers stick around and keep improving.

She contrasts this with advice she got a decade ago to step away from craft and focus on management skills like running one-on-ones, which she calls foreign to her. She credits Shop's designer retention to a leadership style where one-on-ones center on live critique of real prototypes rather than career-development check-ins.

design-leadership-in-the-ai-era

Shopify makes internal decision-making legible by transcribing all meetings and logging every project's proposal, prototype, decision, and rationale in a tool called Vault (GSD), which is hooked up to an MCP so leaders can query past decisions directly through an agent.

She cites this as a practice other design orgs could adopt: instead of relying on institutional memory, a product leader can ask an agent 'have we worked on this before, and where did it fail,' making the org's accumulated judgment queryable rather than lost. Paired with Shopify's data portal, which lets non-technical people query production data directly, she frames these two systems as the biggest unlock for designers moving from 'great craft work' to 'great product thinking.'

ai-prototyping-workflow

Companies

Shopify - Katarina is Design Director for the Shop app; the episode covers Shop's origin, its big product bets, and how Shopify's design org has adopted AI prototyping.
Amazon - Cited as a much larger marketplace that has since gotten inspired by Shop's content-driven, storytelling feed design, which Katarina takes as validation that the bet worked.
Target - Cited alongside Amazon as another large retailer that has moved its discovery feed toward the storytelling-card format Shop pioneered.
Meta (Facebook) - Where Katarina first encountered the prototyping tool Origami roughly 12 years ago, before it was later adopted by talented designers who joined Shop.
Jitter - Mid-episode sponsor read; a design tool whose latest release lets users build custom animation tools inside Jitter by describing effects in natural language.
Desen (dsn) - Mid-episode sponsor read; a tool that extracts a team's design language from its live codebase so anyone can prototype directly in the codebase without opening an IDE.
Notion - Referenced via 'Max from Notion,' cited for making the same argument as Katarina that it's more important for designers to use code to render the fullness of an idea than to worry about whether it ships to production.

Techniques and frameworks

Shoplifting sprints - An annual (informally recurring) two-to-four week 'pencils down' period where a small group of designers goes all-in on polish across every surface of the Shop app; the cart's 'undercart' drawer interaction came out of one of these sprints.
Resetting the baseline / tolerance for risk - Katarina's framework for greenlighting risky redesigns of high-stakes surfaces (e.g., the Shopify signup page): explicitly agreeing upfront that a new design may underperform the incumbent at first, and committing to iterate on it just as much as the original was iterated on, rather than expecting a first-shot win.
Desirability, feasibility, viability mapped to craft - IDEO's three-part innovation framework, which Katarina maps loosely to design (desirability), engineering (feasibility), and product management (viability); she argues the boundaries are blurring but design should remain the tip of the spear for desirability regardless of tool.
Render the idea, don't worry about shipping it - Her core prototyping advice for designers using AI/code tools: focus on rendering the full fidelity of an idea in the most believable medium available, rather than trying to end-to-end ship a production-ready feature; treat the prototype as the artifact you bring to engineering, not the final deliverable.
GSD (Vault) - Shopify's homegrown internal tool that documents every project's proposal, prototype, decision, and outcome, transcribes meetings, and exposes that history to an MCP-connected agent so leaders can query past decisions instead of relitigating them.

Summary

Katarina Batina, Design Director for Shopify's Shop app, joins host Vid to unpack how Shop's biggest product bets got made and how AI has reshaped the way her design team works. She traces Shop's origin to the merger of two internal Shopify projects, a package-tracking tool called Arrive and a checkout product called Shopify Pay, brought together under an "audaciously" named brand with a mandate to become a merchant-first shopping destination. The through-line for most of the conversation is Shop's willingness to reject standard commerce interface conventions, high information density, aggressive micro-optimization, for a "window shopping" model built around storytelling and brand immersion, even when it made product managers and data scientists nervous. She credits that bet's eventual validation to seeing much larger marketplaces like Amazon and Target adopt similar storytelling-card feeds afterward.

A large portion of the episode covers how Shop's designers now prototype with AI. Designer Luke Deon's work reimagining the "shop store" concept is presented as the model: not a single end-to-end feature but a whole configurable world of merchant storefronts, built by talking through the idea philosophically with Claude, developing a plan, and then building against real production data (made safe by developer-experience tooling) and hosted internally for instant sharing. This have led Shop's design org to migrate prototyping work away from React Native toward native Swift and Kotlin, purely because those languages are more legible to both designers and AI coding agents, a technical direction that started as a design-led push and is now forcing Shopify's CTO and founder Toby to reconsider the platform's committed tech stack.

Katarina and Vid dig into the tension between designing without regard for feasibility and the counterargument that code, as a medium, quietly reintroduces feasibility constraints that a rough hand sketch would avoid. She largely agrees with the host's pushback but holds that design's core differentiator is desirability, not implementation, and that designers should render ideas in whatever medium is most believable, using the prototype as a conversation-starter for engineering rather than a production deliverable. She describes "shoplifting" sprints, two-to-four week "pencils down" periods where a small group of designers polishes every surface of the app, as the origin of signature details like the cart's "undercart" drawer, and describes resetting risk tolerance explicitly (as with a redesign of the high-stakes Shopify signup page) as the mechanism that lets big, uncertain bets actually ship.

As AI prototyping produces far more ideas than the team can build, Katarina says editing and disciplined focus, choosing the handful of primitives (package tracking, the feed, agentic search, the Shop Pay identity layer) that actually drive valuable sessions, has become one of the most important parts of her job, replacing the old annual moonshot-planning ritual with continuous, week-to-week idea validation. She also argues that as manual execution time collapses, articulate critique and a real design vocabulary (hierarchy, balance, symmetry) become the main lever for improving work, since feedback to AI agents is now given entirely in words, and many self-taught designers lack that vocabulary.

The conversation closes on leadership and organizational infrastructure. Katarina rejects the older idea that design leaders must step away from craft toward pure management, arguing that staying close to the work is what retains designers and keeps a leader credible in engineering and data conversations. She highlights Shopify's internal tools, Vault (also called GSD), which transcribes meetings and logs every project's decisions with an MCP hook so leaders can query prior decisions through an agent, and a data portal that lets non-technical people query production data directly, as the structural unlocks behind Shop's shift from "great craft work" to "great product thinking." She ends on Toby's adage: figure out what's true, then figure out what to do about it.

Notable Quotes

"Think about a place about a former project or a project you're in now where you feel like you're being blocked and it inherently feels like a different craft's responsibility that's the blocker, and say actually there's nothing stopping me from like jumping the pond here and like embedding myself in that craft or in that skill set or learning it so that I can come to my own conclusions about this thing." - Katarina Batina

"You need to create the grounds to take some risk knowing that you're going to commit just as much iteration into the next thing as you did the the incumbent product." - Katarina Batina

"Focus on rendering your idea because it's literally in the span of time been like a total of 2 hours historically that designers could productively ship to code if they didn't know code in the first place." - Katarina Batina

"Our material language for building is now becoming words. We now have to give critique to our agents. Most the time we have to point at the thing and say, 'This isn't right, and here's why.' The higher fidelity that designers can have in the language they use to make their work better, the better off they're going to be in an agentic era." - Katarina Batina

"Figure out what's true and then figure out what to do about it." - Katarina Batina (quoting Toby Lutke)