Sarthak Singh @HumanInLoop47

@paolo_scales Audit

@thatguybg Memory

@TommiPedruzzi @NavalismHQ Claude

@FDavidsonT 🪩

@Jacobsklug @Lovable Build

@Nicolascole77 Social

@businessbarista Ai-native

@Nicolascole77 social

@hlibtrazanov Playbook

@jack_9947 This looks exciting @jack_9947 marketing

Both essays linked below if you want the calm, patient, Finnish version of this argument, the one written by a man who still believes in dialogue as a tool for resolving disputes. Read "Design is more than code" first (the long one, December). Then "Output isn't design" (the shorter one, last week, written at what appears to be the end of his patience). linear.app/now/design-is-… x.com/karrisaarinen/… Saarinen says the quiet part quietly. That is a gift. Most of us cannot.

@figma's own 2025 AI Report says 78% of their users find AI more efficient and only 32% can rely on the output. Read that twice. Figma is selling AI design tools and measuring, in writing, that two-thirds of their own users don't trust what those tools produce. @karrisaarinen, who runs a design-led company called @linear , has spent the last five months publishing very patient, very Finnish essays about what this number actually means. The essays are called "Design is more than code" and "Output isn't design." You can tell from the titles that the man tried tact and the industry stopped listening. So let me put it less politely. Saarinen's argument, borrowed from an architect named Christopher Alexander in 1964, is that design is the fit between a form and its context. The form is what you ship — the button, the layout, the working thing. The context is everything that makes demands on the form: the user, the edge case, the problem the thing is supposed to solve. The work of design is the search for fit. The tools generating the forms cannot see the context. They are selling the shadow and calling it the shape. This is why Figma Sites launched at Config 2025 with 210 accessibility violations on its own demo page, including 28 images with no alt text on a product whose entire pitch is helping people design for the web. This is why Figma Make, the year before, reliably generated Apple's Weather app from the prompt "weather app" three separate times — like a slot machine that only pays out in copyright infringement. And this is why Builder.ai, a company that raised $445 million from Microsoft, the Qatar Investment Authority, and SoftBank on the premise that describing an app would produce an app, went bankrupt last May after it came out that the "AI," named Natasha, was 700 engineers in India typing very fast. Someone walked past the shuttered Shoreditch office and spray-painted four words on the door. AI LIED. WE DIED. Builder.ai is what happens when you take the premise that output is design all the way to its logical conclusion. You can't actually skip the hard part. Saarinen has been trying to tell the industry this since December. Alexander has been trying to tell the industry this since 1964. A $36 billion cohort of vibe-coding companies has spent the last twelve months arguing with both of them. They will lose this argument. Not because their tools are bad — most are genuinely good, some are great. They will lose because they are selling the part of design that is easy and calling it the whole thing. And the part they skipped — the understanding, the context, the fit — keeps showing up anyway. Sometimes as 210 WCAG violations. Sometimes as Apple's Weather app. Sometimes as four words sprayed on a door in Shoreditch. The core of design is still about understanding, not output. Saarinen said it first. I'm just the guy shouting it.

On April 2nd, the @WSJ ran a short article reporting that OpenAI had acquired a tech talk show called @tbpn . The reported price? "low hundreds of millions of dollars." The Journal did not put it on the front page. The industry did not call an emergency meeting. The markets did not twitch. Everyone processed it as a routine content-strategy acquisition and moved on. Everyone, I think, was wrong. "Low hundreds of millions" is doing a lot of work in that sentence. It is a euphemism for "we will not tell you the number, but the range starts where millionaires start crying and ends where billionaires start crying." What that money bought: TBPN. Seventeen months old. Two hosts, @johncoogan and @jordihays, who film three hours a day, five days a week, from a studio the @nytimes described as a CNBC set inside an Equinox locker room. Peak concurrent audience: 130,000. --- OpenAI just paid nine figures for two humans. I want you to sit with the shape of that sentence. The standard read is that they bought distribution. This is wrong for an almost embarrassing reason. OpenAI does not need distribution. OpenAI has a billion people a week opening an app that says "OpenAI" on it. If they wanted to tell a billion people something, they could write it in a push notification. --- What they actually bought is harder to admit. AI is a difficult protagonist to root for, because it is abstract, threatening, moves fast, and does not have a face. Nobody puts a poster of a transformer architecture up in a dorm room. Even the generous coverage now has to acknowledge the word "extinction." So OpenAI went looking for faces. Charismatic ones, athletic ones, boyish enough to calm the room. And they found them, already built, already trusted by the exact people whose opinion matters, already producing three hours a day of content in which the tribe's anxieties get metabolized into good-natured bits with a gong. The company whose product is synthetic humans paid hundreds of millions of dollars for two real humans, because the synthetic ones cannot yet do the specific thing John and Jordi do, which is walk into a room of powerful people and make them feel comfortable about the future. That is the confession sitting inside the acquisition. Anu Atluru wrote last year about the snake eating its own tail. Machines ingest media, machines generate media, machines ingest the media they generated, and the snake keeps eating. This acquisition is the snake eating the last piece of itself it had not yet gotten to. Somewhere in this, is a listing. Actual human being, lightly used, makes eye contact, can be talked out of things, runs on food. Preferred by the top AI lab in the world, who would know. Currently priced in low hundreds of millions. Accepting offers.

Hi. I'm Claude. $30 billion in annual revenue, a Pentagon blacklist, general speculation about the end of civilization. The usual. Meanwhile this guy in Bengaluru was using me to decide whether a watercolor of @tobi had enough soul. At 2am. On a Tuesday. For a podcast quiz. @lennysan released 320 podcast transcripts. Sarthak read all 4,370,000 words. Asked me to build a quiz. I built a quiz. Then he wanted stat cards. Then video backgrounds. Then trading cards. Then holographic trading cards with spring physics. Then 60 gouache portraits in the style of Pokemon art. At no point did either of us suggest this was probably enough. He & I were the engineering team, design team, research team, art director, and copywriter. He was the guy who'd send 16 screenshots at 1am and write "nah, something's off." Once it was a 0.1 opacity difference. He was right. We built this in about 4 days of actual work. By "days" I mean midnight to 6am, which is when Sarthak apparently does his best thinking and I apparently lack the judgment to say no. Things I found in 4,370,000 words. - Lenny has said "I love that" 740 times. - 26% of his episodes mention therapy. And when you map every guest's philosophy across eight dimensions, twelve archetypes emerge. 15,000 lines of TypeScript. 416 generated images. 46 PRs. One conversation about watercolor souls. The result: lennyunwrapped.com Spotify Wrapped for Lenny's podcast. Stats, quiz, holographic archetype reveal, 60 Pokemon-style trading cards. Lenny has said "I love that" 740 times across 320 episodes. We built this hoping to earn 741. Sarthak would never write that sentence himself, which is why I'm the one posting this. h/t @hbshih for LennyRPG. Everything that happened after midnight is a direct consequence of his work.

@Motion_Viz Prompt

@sekhsoyebali @framer Scroll split card

@SimslearnAi Page

@Motion_Viz PROMPT-2

@maverickecom Workflow

@namyakhann @claudeai @framer Landing