the fake winners

I had a conversation with a friend the other week that i can't stop thinking about.

We were talking about a founder we both know. Guy went through YC. Raised 2 and a half million for a workflow automation platform. Looked like the dream from the outside. Top-tier accelerator, respectable round, clear category, all the signal you could ask for.

And then something interesting happened.

All the "revenue" went away.

Because the revenue wasn't real. It was other YC founders buying into their product. Batch-mates writing checks to batch-mates. A closed loop of validation that looked like PMF and was actually just… people being nice.

They went out into the real world after the batch ended. And the real world said: "no thanks."

The founder didn't have conviction about the space. He'd picked workflow automation because it was a hot category, not because he was obsessed with the problem. So when the fake revenue dried up, there was nothing underneath. No muscle. No taste. No twenty 2am conversations with real users that had taught him the shape of the real problem.

Just a pivot. And then another. And then another.

I want to be careful here. I'm not saying YC is bad. i'm not saying founders in accelerators are frauds. most of them aren't.

What i'm saying is that there's a specific flavor of validation that's killing a lot of founders right now, and almost nobody is naming it.

The category is broader than YC. It's the entire startup ecosystem right now. Accelerator batches where everyone trades logos. Demo days where investors sprint-clap for products they'll never use. SF happy hours where you pitch to three VCs who all text each other afterwards. Twitter threads that go viral within the 2,000-person builder-bro echo chamber.

This isn't traction. This is theater.

And the people performing in the theater often can't tell the difference. Because it FEELS like traction. The metrics go up. The revenue is real-ish. People are paying you. But every paying customer knows your cousin's roommate. Every single "win" traces back to the same tight cluster of people validating each other into oblivion.

And then you raise your next round on those numbers. And now you have obligations built on top of a fiction.

My friend, the co-founder, the builder, told me the story of the workflow automation founder and what struck me was how matter-of-fact he was about it. Like this was just a known pattern. The fake YC revenue trap. Everyone in the ecosystem sees it happen. Nobody writes about it.

So let me write about it.

Here's the meta-problem. It's never been easier to launch.

I wrote about this recently: 41% of code written globally is AI-generated. 557,000 apps hit the App Store last year alone. A solo dev can clone most SaaS products in a weekend. The marginal cost of building v1 has collapsed to nearly zero.

Building is free. Launching is free. Getting the first 100 users is free-ish.

What's still hard, insanely hard, is building something that doesn't die the second the novelty wears off.

And we're entering an era where the signals that USED to tell you "this is working" don't work anymore. Because the same tools that made it easy to build a real product made it easy to build a fake-looking real product. The same networks that used to validate real traction now validate performative traction at scale.

Let me give you three case studies. In order of how much money was burned on each.

Exhibit A: Sora.

OpenAI launched Sora in February 2024 and everyone, and i mean EVERYONE, lost their minds. The demo videos were genuinely incredible. Cinematic. The kind of thing that made you stop scrolling and actually watch.

The product rolled out in December 2024. Shipped to ChatGPT Pro users, then to Plus. Expanded internationally. Got dedicated app treatment. Got memes. Got a full year of being THE hot AI product in the conversation.

Then it got quietly scaled back in early 2026. Limits tightened. Features shuttered. The app became a mostly-inactive appendage to a company that has the most AI eyeballs on earth.

Let me say that again because i think it's important: OpenAI, the company with hundreds of millions of weekly users and more AI distribution than any company in history, could not turn Sora into a durable product.

Not for lack of virality. The virality was insane. People were sharing Sora clips on Twitter for months. The launch announcement alone had more engagement than entire product launches from competitors.

Virality wasn't the problem. The problem was that virality was the ONLY thing. There was no retention underneath. People would generate 3 videos, share one on Twitter, close the tab, and never open it again. It was a party trick. A really impressive party trick with a $200M training cost behind it.

You know who knew this within weeks? Every consumer product person who's ever had to deal with a 95% drop-off rate. Gamma's founders called this out explicitly. Viral tweets and Product Hunt launches "mask a lack of true PMF." Signups spike. Stagnation follows. The first 30 seconds aren't magical, so the 31st second doesn't happen.

Sora had the biggest viral moment in AI consumer history. And it died.

If that doesn't convince you that eyeballs aren't traction, nothing will.

Exhibit B: Openfly. Paperclip. Whatever's next.

These are smaller examples but i want to include them because they're still warm.

Paperclip had its moment. A few days of Twitter virality. Screenshots everywhere. Everyone was gonna use it. Then everyone just… didn't. I haven't seen a Paperclip mention in weeks.

Openfly is getting the same treatment right now. Hot. In the conversation. A lot of people talking about it on Twitter.

But here's what i keep noticing when i actually talk to people: i don't know anyone who uses these pragmatically in their day-to-day. I know people who tried them. I know people who posted about trying them. I know zero people who integrated them into a workflow they'd miss if it disappeared.

This is the crater between "people are talking about your product" and "people are using your product." And right now that crater is enormous. Twitter consumer AI discourse is almost entirely disconnected from actual consumer AI usage. The apps people TALK about and the apps people USE are two completely different lists.

If your only metric is "am i in the conversation on Twitter?" congratulations, you're building a public relations exercise, not a company.

Exhibit C: The Founder's Notes App.

I have a confession.

My notes app on my iPhone has about 150 startup ideas in it. Maybe more. I stopped counting a while ago.

Some of them are genuinely pragmatic, too. Like, there's a Seattle-community version of the LinkedIn boost invite-only thing the SF founders are all doing. I could probably stand that up in a week. People would join. It would make money.

There's another one i think about constantly: a waterfall social listening platform. Nobody has built one. Every social listening tool sucks at one thing or another, so people like me end up duct-taping three of them together. A Clay-style waterfall across existing tools would be extremely useful. I know this because i've personally bootstrapped that duct-tape for two years.

Boring ideas. Real problems. Fast paths to revenue.

And i haven't touched any of them.

Because every single one of them is a trap. Not because the ideas are bad. The ideas are fine. The trap is that any of them could generate JUST enough fake-looking real traction to pull me away from the thing Tyler and i have been chiseling for the last six months.

I'd build the LinkedIn boost thing. It'd get 200 Seattle founders in it. They'd cross-promote each other. The numbers would look amazing. I'd think "huh, maybe THIS is the thing." I'd pivot six months of conviction into a community growth hack.

And two years from now i'd realize the entire thing was a contained ecosystem where a few hundred Seattle founders validated each other into oblivion, and the moment i tried to expand beyond them, nothing happened.

Sound familiar? It's the YC workflow automation story in miniature.

The shiny object isn't the problem. The shiny object is just an object. The problem is that shiny objects produce fake signal fast, and fake signal is indistinguishable from real signal until the moment it isn't.

OK so here's the thing.

I want to talk about someone who's doing this right, because otherwise this whole post is just complaining.

Joe Golden is the guy i keep coming back to. You've probably never heard of him. He spent 14 years co-building Collage.com (2007-2021) before selling it to a private equity firm called PSG, who rolled it up with ShootProof into a new entity called Foreground. Fourteen years. In the consumer photo printing space. While every hot consumer company of that era came and went, Joe sat at Collage and iterated.

After the exit, he started PerfectRec in 2022, a recommendation engine. And in late 2024 he pivoted into a product called paper2audio. It converts academic research papers into listenable audio. Bit of a niche tool. Not trending on Twitter. Not on the front page of Product Hunt this week. Not getting written about by the AI influencers.

Do you know how many users it has?

50,000.

In one year.

Fifty thousand real humans who found the tool, kept coming back, and made it part of their workflow. Two-person team (Joe and his co-founder Chandradeep). Used by researchers at Stanford, Berkeley, MIT, Columbia. Joe also co-teaches a Stanford class called "The Business of AI."

They have a generous free tier. They've gotten the price per hour of transcribed content down to something like 2 cents. They've grown into enterprise deals because when you have 50,000 active users, you start attracting users who have enterprise problems. The business buyer doesn't want their data training OpenAI's models, so they pay for the private tier. That's now a notable revenue line.

They're seed-strapping. That's the term Tyler came up with, it's seed + bootstrapping mashed together. You raise a small amount to get going, then you operate like a bootstrapped company from there on out. No VC treadmill. No "we need to grow 20% month-over-month or we die."

Joe doesn't have "YC batch revenue." He doesn't have "Twitter viral moment" revenue. He has 50,000 real humans choosing his thing over alternatives.

That's the most boring sentence i've written in this entire piece. And it's also the most valuable.

Paper2audio has issues. Joe knows it. People come and go. There's competition. Not everyone wants to listen to podcasts. But he's in year whatever-it-is of just… keeping going. And every year the moat gets a little deeper. Every year the user count goes up. Every year the product learns something the competition hasn't had time to learn yet.

And before paper2audio, he already put 14 years into Collage. The man doesn't quit. That's the whole pattern.

50,000 > 50 million viewers who saw one Sora clip and never came back. Every time. Because the 50,000 are real.

Here's what i keep coming back to in these founder conversations.

The question that torments people is: should i double down or do something new?

And i think the framing is broken. Because "something new" always feels like progress and "double down" always feels like stubbornness. But that's a feeling. Not a fact. And feelings about conviction are the single most unreliable compass a founder has, especially in an era where shiny objects are everywhere.

I wrote about this in the slow software piece. 75% of YC startups pivot before demo day. Which sounds like evidence that pivoting works. But the companies you've actually heard of, the Notions, the Gammas, the Slacks, took 3-5 years of sitting with the same problem, through the claustrophobic part, through the 95% drop-off rate part, through the near-bankruptcy part. They didn't pivot their way to product-market fit. They chiseled their way to it.

The pivoters mostly die. We just don't hear about them because they died quietly. The survivors who pivoted early get survivorship bias'd into "pivoting is the right move." And meanwhile the founders who actually won just sat with the problem for an unreasonable amount of time.

Here's the test i've started applying to myself.

If the signal i'm looking at can ONLY come from people who look like me, went to the same schools as me, live in the same cities as me, know the same founders as me, that's not signal. That's a mirror. Mirrors can't tell you if your product works. They can only tell you if your product is acceptable to your tribe.

If the signal comes from someone who has no reason to care about me. A Mercer Island mom filling out enrollment forms, a 14-year-old with 30 seconds of patience, a random researcher at ETH Zurich who just needs a paper read out loud on their commute. THAT'S signal. That's the real world. That's someone who had to actively choose your thing over everything else competing for their attention and their money.

YC revenue is a mirror. Twitter virality is a mirror. Seattle founder-community mutual promotion is a mirror.

50,000 randos who keep opening paper2audio every week because they actually want to listen to papers? That's a window.

You need the window. You cannot build a company on a mirror.

I'm 22. And honestly, i feel every month of my life carrying a lot of weight right now.

Which sounds dramatic. It kind of is. But it's also true. Two years ago i was fucking around in college. Now i'm six months into building something Tyler and i both want to be here in five years. And i can feel the compound curve of each decision.

Every month i spend on the real thing is a month of context accumulation, of taste development, of learning the problem at a depth that can't be faked. Every month i spend on a shiny object is a month off that compound curve. And you don't get the months back.

There's an argument people make that "you have time, you're young, you can experiment." And it's partly true. But the people who made it biggest usually made it earliest, before marriage, before kids, before the house, before the default life overhead that makes pivoting expensive in a way it wasn't when you were 22 and could sleep on a couch.

So the window where experimenting is cheap is actually smaller than it looks. And every fake-winning shiny object spent during that window is a compounding loss.

I wrote a few weeks ago about how the funding ladder is breaking. Chris Neumann's observation that seed VCs are acting like pre-seed VCs. The collapsed cost of building means you don't need to raise to explore anymore. You can just build. And if you can just build, then the real test isn't "can i get funded" it's "can i get paid by people who don't know me."

That test is simple. And brutal. And most founders never actually take it. Because the fake validation economy is SO good at providing alternative forms of success signal that you can go years believing you're winning without ever putting your product in front of a stranger who has to choose it on the merits.

I'll tell you what Tyler and i are doing, because i don't want this to be abstract.

We're not chasing. We have our 150 shiny objects. We're not building them.

We're building the thing we started building in September. Only deeper. Only more refined. Only 5% better today than yesterday, which compounds to unrecognizable over 180 days.

We're going to make money on the thing that already works. Not because it's sexy, but because making money means we earn the right to experiment. You can't afford to take real shots if you don't have a base. Boring businesses fund interesting bets. Interesting bets without boring businesses underneath them are just prayers.

Tyler said something in a conversation recently that stuck with me. Something like: "step one is make money. Then we can afford to have some experiments alongside the main thing. The best scenario is you have time to play around with what you think is really important, but you also have the structured time to make money."

Money isn't the goal. Money is the permission slip.

And the permission slip comes from boring. From sticking. From not doing the other 150 ideas. From posting twice a week even when it's hard. From showing up every day to the same problem and letting the compound curve do its thing.

If you're reading this and you've got a product that's somehow "working" but every paying customer is someone you went to school with, or met at a demo day, or who writes you checks because they're in your group chat:

Get out of the building. Go find a stranger. Sell it to them without using your network. No warm intro. No batch discount. No mutual connection.

If they pay, you have a company.

If they don't, you have a theater production. And no amount of YC revenue, Twitter virality, or investor enthusiasm is going to change that.

Joe's 50,000 paper2audio users didn't come from his network. He built a thing, they found it, they kept coming back. Before that, his 14 years at Collage didn't either. That's the whole trick.

Sora had a billion eyeballs and couldn't make it work. Paperclip had its moment and evaporated. A generation of YC batch-mates are currently paying each other for products nobody outside their cohort will ever use.

And somewhere out there is a paper-to-audio tool with 50,000 real users who genuinely need it, operated by a guy who spent fourteen years on his last company, quietly building something that will still exist in a decade.

That's the game. Everything else is performance.

-parsa