Product Craft and the Role of Design

This is a narrative theme, with a typological spine, tracing how craft and taste function as competitive moats in software products — and how the role of design is expanding from execution toward judgment. Synthesised from Dylan Field on Figma and Product Taste, Dylan Field 2.0 on Figma, Jenny Wen on Designing Claude, the Legibility Framework, and the Three Designer Archetypes, Karri Saarinen on Linear, the Linear Method, and Craft-Driven Product Development, Katie Dill on Design Quality, the ROI of Design, and Operationalising Excellence, Brian Chesky on Airbnb and Product, Scott Belsky on the First Mile, Doing Half, and the Messy Middle, Nickey Skarstad on Vision-to-Execution, Product Quality, and Second-Order Thinking, Mihika Kapoor on Figma, Zero-to-One Products, and Democratic Meetings, Max Schoening on Malleable Software, Agency, and Building Products in the AI Era, Stewart Butterfield on Product Craft, Utility Curves, and Why We Don't Sell Saddles, and Nan Yu on Speed, IC-First Product Design, and the Extreme Version Method.

The core claim

As features commoditise and building cost collapses, the felt quality of a product — the coherence of its interactions, the care visible in its first thirty seconds, the discipline of what was left out — becomes the durable differentiator. Craft is not finishing polish added at the end; it is a structural decision made at every stage of building, encoded in what the team chooses to ship and what it chooses to kill.

The role of design is shifting to match. Where design once operated as a service function — making things look good after strategy had decided what to build — it now sits at the upstream decision: what is worth making, what form should it take, what standard of quality is non-negotiable. Dylan Field states the position directly: ‘Good enough is not enough. It’s mediocre.’ Jenny Wen frames the same shift as design stratifying into two distinct activities — execution support, which AI compresses, and vision creation, which it does not.

Craft as moat, not decoration

Stripe documented a 10.5% increase in user revenue attributable to accumulated checkout-flow quality improvements — an unusually large lift from small design decisions compounding over time. Katie Dill draws the argument from this: beauty increases trust in B2B, because a polished checkout signals to a merchant that Stripe will not surprise them in production. Design quality is not aesthetics separate from function; it is a trust signal that converts at the boundary between unfamiliar parties.

Karri Saarinen arrives at the same position from the other direction. Linear built a product widely considered best-in-class in quality with roughly fifty people, profitable, with negative net burn. The mechanism is not a dedicated design-review process but a hiring bar: only people who feel personal ownership over what ships. Quality cannot be procedurally imposed on people who do not intrinsically value it. Linear tests for Product Taste in engineering interviews by asking candidates to evaluate real product decisions — ‘Why was this decision made? Do you agree with it?’ — because taste cannot be taught as easily as technical skill.

Brian Chesky operationalises this at the leadership level. His CEO review cadence — reviewing all product and marketing work directly, on a rotating schedule — was designed precisely to hold quality without bureaucracy. ‘There’s a difference between micromanagement, which is telling people exactly what to do, and being in the details.’ Being in the details is what Friction Log exercises formalise at Stripe: designated people walk the product as specific personas and log every point of friction and delight, feeding a quarterly Product Quality Review.

The first mile and what teams neglect

Scott Belsky names the asymmetry cleanly: ‘Most teams spend the final mile of their time building the product, considering the first mile of the customer’s experience using the product.’ New users enter in a specific psychological state — lazy, vain, selfish — and no amount of feature depth compensates for a first thirty seconds that makes them feel stupid or incapable. The First Mile Experience is the most neglected design surface precisely because the team’s own users are not in that state.

Stewart Butterfield makes the parallel point about comprehension versus friction. Reducing clicks is almost always the wrong goal. Reducing the amount of thinking required is the right one. Most users arrive at a new product just above the threshold of intent, with near-zero specificity about what they want. The design question is whether someone can look at a screen, understand what the product does, and understand what to do next — not whether they can reach it in two taps.

Nickey Skarstad extends this to quality metrics. At Airbnb Experiences, review rate was the north star precisely because it encoded what the team actually cared about: experiences worth having, not just bookings. When a feature drove more bookings but worse reviews, the team did not ship it. A growth metric without a quality constraint optimises toward a product that is large and hollow.

Taste: what it is and where it comes from

Product Taste is the capacity to judge which of many possible things is worth building, how to build it well, and what a delightful experience looks like. Multiple practitioners in the wiki treat it as learnable but not teachable by instruction alone.

Dylan Field defines product intuition as a hypothesis generator: ‘You’re constantly generating these hypotheses… You put them forward, debate them, try to find data to support or negate them, then winnow it down into a working hypothesis.’ Taste is not an oracle that delivers verdicts; it generates candidates for testing. The practical expression at Figma: furrowing his brow at complexity and insisting there must be something simpler — not a final answer, but a forcing function for a specific question.

Max Schoening offers the most operationally precise model: taste is a simulation capability. Given an idea, a person with calibrated taste can run a simulation — for this specific in-group, in this specific context, will this idea resonate? The distinction matters: ‘I like this’ is not taste; ‘my target in-group will find this excellent’ is. Taste built this way is trackable — you can test predictions and update them.

Dylan Field’s fullest account of how taste is built is a five-step loop: have an experience; react to it; build the canon (understand the path of decisions that led here); agree or disagree philosophically with that path; repeat across domains and find cross-domain correlations. Most people who do this consistently learn to match a framework — understand a genre and execute within it with high fidelity. The rare few learn to create the framework. Both matter: framework-creators set direction, framework-matchers execute.

Nan Yu locates taste at the execution level, not the strategic one. Her IC-first philosophy holds that the people closest to the pixels are also best positioned to judge whether something is right. Moving to management abstracts leaders away from that judgement. Her Extreme Version method is taste made operational: build the strongest possible version of an idea first, because attenuated versions leave open the question of whether the idea was right or the execution was weak.

The typological spine: three contested axes

Design as service vs. design as leadership

The conventional model positions design as a service function — catching things before they go out the door, making them look better. Jenny Wen frames the alternative: the premium has shifted toward vision creation, deciding what should exist and what the interaction model should be. AI compresses implementation time; the designers who survive the shift are those who moved upstream before compression made execution cheap.

Brian Chesky made a structural version of this move at Airbnb. When the design community cheered at Figma Config on hearing Airbnb had restructured PM, what they were reacting to was a Silicon Valley norm of treating design as a service organisation rather than a development partner. The restructuring embedded design in the product process from the start rather than bringing it in to polish output.

Taste as innate vs. cultivated

The received view of taste treats it as a quality some people have and others do not — a function of temperament or background, not trainable. Every account in the wiki argues the opposite. Guillermo Rauch runs ‘exposure hours’ as an internal operating principle at Vercel: quantify how much time you spend watching real users interact with your products, because the inertia is always toward thinking you know how the product works. Demo Fridays, customer meetings where Rauch uses customers’ products live rather than discussing features abstractly, and inviting customers to demo their usage in front of the whole company — all are deliberate taste-training mechanisms.

Jenny Wen’s Three Designer Archetypes make the same argument in hiring form: the block-shaped generalist, the deep-T specialist, and the craft new-grad each represent a different shape of adaptability, not a fixed level of innate ability. The shared baseline is resilience and willingness to drop old methods when the craft is changing faster than any fixed skill set can keep up.

Craft as slowing down vs. craft via faster iteration

An intuition persists that quality requires more time — more review cycles, more deliberation, more senior oversight. Karri Saarinen inverts this: the craft-driven path runs through faster internal shipping, not thoroughness of pre-release review. Visible internally only → feedback from one to five customers → polished general release. Over-investing in polish before external contact wastes resources that should go into the iteration cycle. The feedback loop is the quality mechanism, not the pre-release review.

Nan Yu makes the same argument from operating clarity: when everyone understands the product standard and trusts the standard, decision time collapses. Deliberation happens once, when the standard is set, not repeatedly for each feature. Speed and quality are not a trade-off; the belief that they are usually disguises a lack of conviction about what good looks like.

Dylan Field holds the limit case from the opposite direction: Figma took 3.5 years to launch and five years to a paying customer — and his conclusion was ‘get it out as fast as you possibly can.’ The craft standard is not about time spent; it is about what ships. The ‘minimally awesome product’ is the smallest scope that is genuinely great for its scope, shipped as soon as it is genuinely great, not sooner.

Design and AI: what changes, what does not

Scott Belsky expects AI to generate the centre — the average, the expected, the optimised — and argues this will liberate people to focus on the non-scalable edge, the craft, the things that move the needle for experience. He is long the experience economy as AI handles production.

Jenny Wen adds texture from inside Anthropic. Design is stratifying: execution support (implementation, component consistency, shipping fast with AI tooling) is being compressed; vision creation (what should exist, what the interaction model should be) is not. The designers who remain essential are those holding the quality bar upstream — which is precisely where the work that AI cannot do tends to live.

Max Schoening frames Malleable Software as the design challenge that emerges from this: software that adapts to the user’s intent rather than locking the user into the designer’s mental model. When AI can interpret intent expressed in natural language, the design constraint shifts from ‘how do we surface the feature?’ to ‘how do we understand what the user actually wants?’ Understanding intent is now the hard problem; implementing intent is not.

The implication for Product Taste: when building the first version is cheap, the value of accurate pre-simulation (taste) increases. Which of the countless things that are now buildable should be built? That question is more durable than the question of whether the team can build it.