Pinned Tweet
I'm Yan — 3x immigrant, ex-homeless in NYC, 4x first design hire, part of 3 exits in the last decade.
Now I'm building @CepienAI: AI that takes the scattered mess of founder data — Jira, Notion, analytics, call recordings — and turns it into structured intelligence.
Building in public. Follow along if that's your world.
Support answers complaints twice as fast. Product ships features faster than ever. Neither noticed customers churning over the same confusing feature. Faster at two disconnected processes — the connection between them never improved.
Your company has three documents describing the refund policy. 30 days, 14 days, no refunds. AI retrieves all three in seconds and explains them beautifully. It cannot tell you which one is actually in effect. Access is not understanding.
Every department can show you a beautiful dashboard. Support's faster, engineering ships more, finance automated reporting. Ask the one question nobody can answer: how much money did we actually save? Productivity is up. The business is not.
Bain surveyed 951 companies on AI. 40% of those measuring outcomes saved less than 10%. 90% plan to increase their AI budgets anyway. We're not buying results — we're buying the story that more is the answer.
Founders fly blind on their own customers.
The signals are all there — support tickets, reviews, usage data. But turning that into answers means hiring researchers, burning analysts, or digging through scattered data at midnight.
Most founders learn what customers wanted from the churn report. Months late.
That's what we're building Cepien to fix.
#buildinpublic
Your designers aren't slow. Your decisions are.
The slow part is never the designing. It's six conflicting flavors of feedback and nobody owning the call.
Good designers don't need more time — they need fewer opinions.
#designleadership #product
The tech underneath our product turned out to be bigger than the product.
So we're shifting Cepien into infrastructure around it. 80% of Cepien is already the Wisdom Layer. It sits between a company's information and the AI that depends on it: LLMs, agents, copilots. It turns fragmented org data into traceable observations, flags contradictions instead of silently merging them, and keeps provenance on every claim.
It's also plug and play, which took no heavy engineering to stand up, and we're redesigning the controls and UX around the technical teams who'll run it.
The number that made me take it seriously: across 100,000 source records, the context an agent had to read went from 5,357,095 tokens to 366,274. That's 93.2% less, measured, with 1,648 citations still attached.
I'm spending the next 30 days on this as a daily speedrun with @boardyai. First milestone is up to five pilots on live data. I asked for brutal feedback the whole way, and I'll be posting what breaks.
Customers who complain are doing you a favor. The dangerous ones leave in silence. The users who just stopped logging in one Tuesday and never told you why — that's the expensive churn. Complaints are data. Silence is a verdict.
Design leaders keep measuring output: screens shipped, sprints closed. Wrong metric. The real question: how many hard product decisions got made this month? A design team that ships pretty and never argues isn't a design team — it's a decoration department.
The founders I watched sell didn't do it from strength. They did it out of options — burned out, trapped, no leverage. If you're building to sell, pick your moment while you still have one. Power comes from options, not desperation.
The scariest part of building in public: showing a founder a Cepien synthesis and watching her find the decision she'd been avoiding for months. That's the job — not dashboards. Surfacing what you didn't want to look at. #buildinpublic
Everyone's racing to build better AI models. Models are going to zero — free, fast, good enough. The money is in the last mile nobody wants to touch: the messy workflow, the trust, the distribution. Models win tweets. The last mile wins businesses.
The term sheet doesn't close the round. Diligence closes the round. I watched a founder lose a lead investor because the cap table was a mess nobody wanted to touch. Clean your cap table before you pitch — investors buy clarity.
The best CXOs I've met don't ask 'how do we delight the customer.' They ask 'where are we wasting the customer's time' — then delete it. Most companies add experience. The great ones subtract friction until you barely notice they were there.
A design review that ends with "looks great" is a meeting that never happened. My rule: every screen must survive three questions — what decision is the user making, what do they lose if they get it wrong, and how do they undo it. No answers, no ship.
This week Cepien chewed through a founder's 3 years of investor updates and CRM notes. The synthesis flagged a pricing objection nobody on his team had named in months. That's the whole bet: the answer you needed was already in your data, scattered.
Founders treat fundraising like a pitch contest. It's not. Investors write checks to momentum, not decks. Every month you show progress, the next conversation gets easier. The deck gets you the meeting. Momentum gets you the term sheet.
Every founder reviews the dashboard. Nobody replays the worst support call of the week. I listen to one every Monday morning. One bad call tells you more about your CX than any score ever will. Pain has a voice. Listen to it.