CoFoun — an AI co-founder, for founders, everywhere
- Brand
- Product design
- Engineering
- AI
Brand, product, AI agent, twelve cities — shipped in eight weeks. The complete v1 of an operating system for founders, on a single deploy.
12
Cities at launch
8 weeks
Brand to launch
70%
Week-one AI engagement
The shape of the problem
LinkedIn turned founder networking into a directory of "open to opportunities" badges and recruiter spam. Twitter/X is a feed where the loudest founder wins. Slack groups go quiet by month two. None of these things were built for founders, and none of them know whose Tuesday meetup is actually worth going to in São Paulo, or which pre-seed in Lagos just closed at terms relevant to yours.
CoFoun started with that gap and a one-line summary: an AI co-founder, for founders, everywhere. Not LinkedIn for startups. Not a community app. Not US-only. Not loud. Built calm, built global from launch, built around an AI that knows the founder's scene — Brooklyn or Bengaluru, Istanbul or Lima.
Eight weeks later, twelve cities went live in a single deploy.
What we shipped
A brand. A product. An AI. A press kit. A weekly digest pipeline. A scene-priming layer for twelve launch cities. The complete v1 of an operating system for founders — front to back, in eight weeks, fixed fee.
The brand, set with care
The wordmark is one word, set in Poppins Medium with −3 tracking. No icon. No symbol. Nothing to redraw. The 18-page brand guide enforces the discipline that became the whole product: direct over polite, specific over abstract, calm over loud.
Color is intentional. Brand Black (#1A1A18) does the heavy lifting at 60% of every surface. Cream (#F7F7F5) softens the web. Pure White carries email and docs. AI Blue (#378ADD) is reserved — single signal, single function: AI is here. Sparkles icon, AI Blue, never anywhere else. Use it for favorites or highlights and the meaning collapses.
Typography is one typeface (Poppins, system-ui fallback), six sizes, three weights. Sentence case everywhere. Never Title Case. Tabular numerals for tables, proportional for prose. Every choice is in the guide because every choice is enforceable in code.
The product, calm by default
The app has four surfaces — profile, map, feed, AI — and one rule: never optimize for engagement. No streaks. No notification by default. No karma, no upvotes, no daily-active gamification. Founders have enough noise. CoFoun's job is to subtract noise, not add it.
The profile is where the AI gets context: stage, vertical, city, what you're building, what you're stuck on. Profile fields are written in the voice founders use — pre-seed, term sheets, design partners, ARR — not the synthetic vocabulary of LinkedIn job titles.
The map shows the global founder graph, weighted by city density and recency. Open it in São Paulo and you see everyone shipping near you. Open it in Lagos and you see the same density, in the same UI, with the same filters. The map adapts to the scene without explaining itself.
The feed is short and surfaced. Three founders nearby raised this week. Two events worth your time. One application deadline that fits your stage. That's the day. The feed ends when the useful information ends — there is no infinite scroll.
The AI lives in every screen. It speaks in the brand voice: direct, confident, specific, founder-fluent. Asked which event to skip Tuesday, it answers with names and a one-sentence reason. Asked to review a YC application, it returns a critique that names the section to rewrite first. Vagueness is engineered out of the prompt layer.
The AI co-founder, built on context
Under the hood the AI is a retrieval-augmented agent with three memory layers — global founder graph, city-scene priming, and per-founder profile — composed at request time and answered through Claude with a tightly tuned system prompt that enforces the voice.
The scene-priming layer is the work most founders never see and the work that makes the AI feel local. Lagos pre-seed expects different signal than Brooklyn pre-seed. Istanbul climate-tech speaks a different vocabulary than Berlin climate-tech. The agent is given the city's idiom and the city's actors before it answers anything. The compute cost is small; the answer-quality difference is the difference between "founder said it's useful" and "founder told another founder."
The vocabulary is founder-fluent in any language. Term sheet, design partner, pre-seed, ARR — these translate, but the meaning shifts subtly per market. We encoded the subtleties. The agent knows that a "design partner" in São Paulo usually means an early enterprise contract and in Brooklyn usually means a co-design relationship. It answers accordingly.
How it shipped
Three roles inside MYC: brand lead, product designer, engineering partner. One AI engineer brought in for the agent layer. Eight weeks, one fixed fee, written SOW, no time-and-materials gray area.
Week 1–2 — Brand. Wordmark construction (cap-height geometry, X-unit clear space, the "don't" matrix), the color system with WCAG-tested pairings, the 18-page guide. Every decision codified in tokens before product touched a screen.
Week 3–5 — Product surfaces. Profile, map, feed, weekly digest. Component library built against the brand tokens; the design system is the brand guide rendered in code.
Week 6–7 — AI agent. Retrieval graph, scene-priming layer, founder-fluent vocabulary, voice-enforcing system prompt. Eval suite for tone (direct, confident, specific, founder-fluent — all four scored before deploy).
Week 8 — Twelve city launch. Press kit, social templates rendered straight from the brand guide, weekly digest scheduling on Resend, Cal.com integration for founder-to-founder intros. The launch went out without a staged tweet — the AI was already useful.
The stack is the one we use for every engagement, plus one layer: Next.js, TypeScript, Postgres, Vercel, Stripe, Resend, Cal.com integration, Claude as the AI engine. Same opinionated foundation. Same fixed fee. Same six-week shape, extended by two for the AI layer.
The outcome
Twelve cities live at launch — New York, London, Berlin, Istanbul, Tokyo, São Paulo, Lagos, Bengaluru, Singapore, Mexico City, Tel Aviv, Seoul. The architecture is built so the thirteenth city is a deploy, not a rewrite.
Five thousand founders in the first cohort, by invite. Week-one engagement on the AI: 70%. Industry average for B2B SaaS week-one engagement: about 25%. The dominant qualitative signal in the first month was a pattern: founders quoted the AI to other founders. "It told me to skip the event. It was right." — Maya, founder, Lisbon. That sentence ran on the testimonial card, exactly as a founder said it.
The brand is what made the rest possible. A typeface decision. A color reservation. A voice rule. Every product surface inherits those constraints — every notification, every digest, every reply. When the AI answers a question, the answer is in the brand. When the email digest lands, it is in the brand. The brand is not a layer on top of CoFoun. It is the shape of the thing.
Calm by default. Global from day one. Founder-fluent in any language. The AI co-founder you didn't know you'd already need.
“MYC built CoFoun in eight weeks — brand, product, AI agent, twelve cities. Founders started asking the AI before we'd even finished the press kit. The voice in every reply is the voice we wrote down on day one.”