All podcasts / Dive Club / Summary

Jason Yuan - Designing 99 "No"s for a single "Yes"

2026-07-15 - 57 min - source - Read full transcript
Rid (host)Jason Yuan

Key insights

Every shipped feature is the distilled residue of dozens of rejected explorations.
Yuan frames his entire design practice around the idea that what ships is only the surviving 1% of extensive divergent exploration - the goal after generating many directions is finding 'the most true and simple expression' of the idea, not the first workable one.
creative-process
Taste is defined as the act of saying no, not having good ideas.
Yuan argues taste is judgment applied through subtraction: liking everything means having no taste, because taste requires actively rejecting most options. This reframes taste as an editing skill rather than a generative one.
taste-and-editing
AI models default to adding or patching, not cutting, which works against good design.
Yuan observes that when you give current AI models critique, they respond by adding fixes or extra elements rather than removing what doesn't work. He attributes this to post-training incentives and says the human designer's real job - editing away - is something the model has no opinion on doing itself.
ai-and-slop
Dot's memory feature evolved from literal fact-bubbles into full narrative editorials before landing on 'Chronicles' as its most legible form.
Early Dot showed memories as Sims-style bubbles listing single facts ('Dot learned that Hermes is a cat'). Over time this grew into structured cards, then 'top of mind' and 'guides' concepts, before the team distilled the underlying idea into Chronicles - described as 'Wikipedia entries of your life' - as the simplest version that was good enough to ship.
product-evolution-dot-to-hivemind
Dot's 2023 'dailies/guides' concept anticipated ChatGPT Pulse by roughly two years but was killed because the technology wasn't ready.
The feature would have had an AI agent do overnight research and present personalized editorial-style guides. Yuan says image and text generation quality in 2023 (GPT-4, 8K context) wasn't reliable enough at scale, so despite building a working version, the team cut it - a decision he says was validated when ChatGPT later shipped something nearly identical as Pulse.
product-evolution-dot-to-hivemind
The name 'Hivemind' won out over alternatives because it was instantly legible to strangers, unlike prior candidates.
Over two years the concept was called 'takes,' 'seniors' (a Brian Eno term for collective genius), and 'small talk' before landing on Hivemind - chosen specifically because it was the only name that let people immediately grasp the idea on first hearing, even though Yuan initially resisted it for being 'too obvious' (bees, hives, yellow).
product-evolution-dot-to-hivemind
The Hivemind asterisk logo was chosen because it visually encodes censorship, emphasis, and incompleteness simultaneously.
Yuan selected the asterisk as Hivemind's mark because it is 'the language of censorship' (used to redact words), a typographic signal for emphasis (bolding), and implies 'there's more to the story' - directly mirroring the product's premise that there is a public story and a hidden, fuller one underneath it.
visual-identity-design
Unstructured social prototypes function as a substitute for traditional user research.
Yuan describes deploying a group iMessage bot ('lore keeper') at events like a South Park Commons demo fair and in New York City, treating people's unscripted behavior with it as data. He explicitly contrasts this with formal user research, arguing 'users don't know what they want' and that giving people situations to react to reveals more than asking them directly.
creative-process
People are more willing to talk about others than about themselves, which shaped Hivemind's core mechanic.
The unexpected finding from the lore keeper prototypes was that people who avoid talking about themselves will readily talk about other people. This observation directly informed Hivemind's design as a system that aggregates many people's accounts of each other into a collective narrative.
product-evolution-dot-to-hivemind
Adjacent creative work (like scoring a launch film) can reveal a product's underlying character before language does.
Yuan describes realizing, while scoring Hivemind's launch film, that the music needed to be improvisational jazz - and only afterward understood why: jazz is musicians sharing one improvised reality, and its noir/detective connotations clarified that Hivemind's personality is a 'truth seeker' like Sherlock Holmes, not a friend or coach.
creative-process
Yuan argues that oversanitized, 'virtuous' design is itself a form of slop because it strips out the full range of human experience.
He criticizes high design's tendency toward a sexless, edge-free aesthetic that removes envy, lust, and vice from consideration, arguing this erasure produces 'essentially nothing' rather than something more honest - and that genuine coherence requires including the parts of human experience that are uncomfortable, not just the virtuous ones.
ai-and-slop
Hivemind's stated north star is giving people an honest signal of what others actually think, distinct from algorithmically curated feeds.
Yuan positions Hivemind against platforms like Google/social feeds that organize and serve public content, arguing that in an age of infinite generated content what people actually hunger for is 'honest and truth signal' and shared narrative, which is the problem Hivemind is built to solve.
ai-and-slop

Books referenced

Media referenced

Companies

Techniques and frameworks

Summary

This episode of Dive Club is a visual walkthrough of designer Jason Yuan's multi-year creative arc, from his previous company New Computer's product Dot (2023-2024) through to his current company, Hivemind. Rather than a conventional interview, most of the episode has Yuan narrating screenshots of concept art, UI explorations, and mood boards that never shipped, illustrating his core philosophy: "every single thing we shipped is the distillation of the 99 nos that preceded it."

The first half traces Dot's design evolution. Early memory UI was directly modeled on The Sims, showing single-fact "bubbles" whenever the AI registered something to remember. This grew into structured cards, a "deep dot" theory-of-mind concept, and eventually "dailies" or "guides" - AI-generated personalized editorials meant to run overnight research and present it beautifully. Yuan says the team built a working version of this in 2023 but killed it because the underlying image and text generation technology wasn't yet reliable at scale; he notes ChatGPT later shipped something nearly identical as Pulse. The surviving distillation of all this work was Chronicles, described as "Wikipedia entries of your life," which Rid singles out as a favorite feature he personally used to relive memories.

The second half covers the pivot from Dot to Hivemind. After shipping Dot, Yuan rejected the "personal assistant" direction (arguing most people, unlike tech workers, aren't bottlenecked on busy-ness) in favor of "connecting the dots" between people. This led to a series of social prototypes - a group iMessage bot called "lore keeper" tested at a South Park Commons demo fair and later in New York City - which Yuan treats as informal social experiments rather than user research. The key finding: people who won't talk about themselves will readily talk about other people, which became Hivemind's central mechanic of aggregating many people's stories about each other.

Yuan then walks through Hivemind's visual identity development in detail: a "cinematic universe" of AI-generated mood images built around birds and murmuration (starling flocking as a metaphor for high-bandwidth collective communication), followed by a deliberate turn toward taboo, occult, and secret-society imagery (Madonna's Sex book, Alexander McQueen x Puma, Blair Witch Project sigils, The Social Register, Gutenberg Bible black-letter typography). The eventual logo - an asterisk - was chosen because it encodes censorship, typographic emphasis, and the implication that "there's more to the story," directly mirroring the product's premise of a public story and a hidden fuller one.

The conversation closes on broader reflections about taste and AI. Yuan defines taste as the act of subtraction and judgment ("taste is just saying no"), and argues current AI models are structurally biased toward addition and patching rather than cutting, which works against good design. He extends this into a critique of contemporary high design as increasingly "sexless" and overedited to the point of excluding real human experience (envy, lust, vice), calling that erasure its own form of slop. He closes by framing Hivemind's north star as giving people an honest signal of what others actually think, distinct from algorithmically curated feeds, and teases a larger reveal still to come.

Notable Quotes

"Every single thing we shipped is the distillation of the 99 nos that preceded it." - Jason Yuan

"Taste is just saying no. Taste is just judgment. It's going like I like this, I don't like this. And if you liked everything, then you have no taste." - Jason Yuan

"The design process is really about editing away... but the model doesn't seem to want to have an opinion on what to edit down." - Jason Yuan

"I don't like making things that are obvious. I like to make the unobvious obvious." - Jason Yuan

"The ultimate north star of hive mind is to give humanity access to a point of view that is honest about what people really think about things." - Jason Yuan