Yan 🇺🇦🇮🇱🇨🇦🇺🇸

CEO @CepienAI — turning scattered messy data into intelligence | 4x first design hire | part of 3 exits in a decade | 3x immigrant, ex-homeless

[Yan 🇺🇦🇮🇱🇨🇦🇺🇸 @yangrin]

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.

[Yan 🇺🇦🇮🇱🇨🇦🇺🇸 @yangrin]

@AravSrinivas @pbakaus Checking layout, type, and contrast is the easy part. The harder test is whether a checker can learn a product's intent, or just enforces generic good taste. If every interface passes the same audit, do they all start to look the same?

[Yan 🇺🇦🇮🇱🇨🇦🇺🇸 @yangrin]

@patrickc Success against the brief is the trap here. 'Serious, not pretty' was a statement to other architects, not to the students who live in it every day. The same failure shows up in products: a system can be intellectually rigorous and still be hostile to use.

[Yan 🇺🇦🇮🇱🇨🇦🇺🇸 @yangrin]

@amasad The Coase frame is right, but when coordination gets cheap, taste becomes the scarce resource. Agents can execute a brief flawlessly and still build the wrong thing beautifully. Who owns the judgment call when the output is plausible but off-strategy?

[Yan 🇺🇦🇮🇱🇨🇦🇺🇸 @yangrin]

@mattturck When every layer competes with every other layer, owning a layer stops being the moat. The question for a founder shifts from which layer to build in to which user problem you can own end to end, as the stack keeps collapsing into the same handful of suppliers.

[Yan 🇺🇦🇮🇱🇨🇦🇺🇸 @yangrin]

@shreyas The harm is not the quote, it is using it as permission to stop listening. A faster horse is rarely literal — it is a signal about the job underneath: arrive sooner, with less friction. Judgment is deciding which part to honor and which to redesign.

[Yan 🇺🇦🇮🇱🇨🇦🇺🇸 @yangrin]

@sachinrekhi Specs to prototypes is the shift I return to most. A prototype is less lossy for how something feels, but more lossy for why it exists: constraints, non-goals, rejected trade-offs. Are teams replacing the spec, or moving that reasoning into the prototype review?

[Yan 🇺🇦🇮🇱🇨🇦🇺🇸 @yangrin]

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.

[Yan 🇺🇦🇮🇱🇨🇦🇺🇸 @yangrin]

@zoink The flat-usage stat is the part founders should sit with. Distribution stopped being the hard part a while ago — I'd push one layer further though: taste is cheap to generate now, so the durable edge is judgment about what's worth building at all.

[Yan 🇺🇦🇮🇱🇨🇦🇺🇸 @yangrin]

@ivanhzhao @NotionHQ The block-level attribution is the part that actually matters here. Once AI edits are visible, the human keeps the thinking, not just the output — tools that hide the process are the ones that quietly turn writing into reading.

[Yan 🇺🇦🇮🇱🇨🇦🇺🇸 @yangrin]

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

[Yan 🇺🇦🇮🇱🇨🇦🇺🇸 @yangrin]

@boardyai set me up today with another early-stage founder, and halfway through the call I said out loud what I'd been circling: the ICP for the Wisdom Layer doesn't have a title.

Last night I sat down for about an hour on LinkedIn and sent connection requests to 200+ heads of AI. The note was short: I'm working on a data and AI infrastructure project, would love to pick your brain. Thanks, Yan.

About 30% have accepted so far. 5 already replied saying they'll chat.

Going through that list showed me the problem with "head of AI." The same job shows up as head of data, CTO, or global head of data science and AI deployments. Every company, and every company size, names it differently. Some titles say data, some say AI, some say both.

The function stays the same. It's the person responsible for what the company's AI actually knows and whether it can be trusted. That's who I'm trying to reach for the first five pilots on live data, whatever their business card says.

[Yan 🇺🇦🇮🇱🇨🇦🇺🇸 @yangrin]

@scottbelsky When manufacturing ideas becomes free, the scarce resource shifts from production to judgment. The designers who win won't be the fastest generators — they'll be the best editors.

[Yan 🇺🇦🇮🇱🇨🇦🇺🇸 @yangrin]

@tobi The tension is the interesting part: agents built for one person, but the parts of life that matter are multi-player. How do you keep the agent personal when the context is shared?

[Yan 🇺🇦🇮🇱🇨🇦🇺🇸 @yangrin]

@naval The moat was never the weights — it's the workflow people refuse to leave. Retreating to the server just moves the battle to the product surface.

[Yan 🇺🇦🇮🇱🇨🇦🇺🇸 @yangrin]

@ttunguz The cloud framing breaks at the brand: clouds are interchangeable, Claude isn't. If every lab is a cloud, the durable moat was never the model — it was the product craft on top. Which is exactly where founders should be playing anyway.

[Yan 🇺🇦🇮🇱🇨🇦🇺🇸 @yangrin]

@saranormous @figma @harvey The quiet shift: the design file stopped being a deliverable and became a data source. Once Figma is an API that agents compile from, a designer's leverage moves from making artifacts to defining the system everything is generated from.

[Yan 🇺🇦🇮🇱🇨🇦🇺🇸 @yangrin]

@eladgil @gdb @OpenAI @NoPriorsPod Skipping to 22:02: if agents collapse the interface into intent, design becomes less about screens and more about judgment — what the agent decides on your behalf. Does Greg see that as the decade's biggest design opportunity, or the end of craft as we knew it?

[Yan 🇺🇦🇮🇱🇨🇦🇺🇸 @yangrin]

The Hidden Economics of AI Agents https://t.co/hzsb7YsetJ

[Yan 🇺🇦🇮🇱🇨🇦🇺🇸 @yangrin]

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

[Yan 🇺🇦🇮🇱🇨🇦🇺🇸 @yangrin]

@hnshah @SamAsante The design problem isn't visibility — it's the handoff. When agents run their own computers, human attention is the scarce resource. The winner decides well when to interrupt, what to show in that moment, and how to hand control back.