I Got Into YC by Hacking Its AI. Meritocracy Is Dead.

You think Y Combinator picks the best founders. That’s cute. It picks the ones who understand the machine.

Here’s what happened: YC deployed a tool called Paxel. Applicants run a cURL one-liner that installs something on your computer, analyzes every line of code you’ve ever written with a coding agent, compiles a report, and uploads it to YC’s servers. Over 100,000 founders worldwide got scored by this system. One applicant reverse-engineered it — figured out what Paxel was actually measuring — and optimized his application to match. He got in.

The gatekeeper built a machine to judge founders. The founders built a machine to judge the machine.

Let that sink in. The most prestigious startup accelerator in the world — the kingmaker behind Airbnb, Stripe, DoorDash — deployed AI to objectively evaluate founders at scale. And the very first thing a smart applicant did was hack the evaluator. Not build a better product. Not demonstrate deeper insight. Just figure out what the scoring algorithm wanted and feed it back.

This is the story nobody in Silicon Valley wants to tell out loud, because it breaks the illusion that matters most to them: that the system works.

Here’s what you need to understand about how Paxel works. It’s not a black box sitting in a data center somewhere. It runs on YOUR machine. It reads YOUR code. It uses a coding agent to analyze how you build software — your patterns, your decisions, your technical fingerprint. Then it packages all of that into a report and ships it to YC.

That’s not evaluation. That’s surveillance dressed up as meritocracy.

Now, to be fair — some commenters pointed out that Paxel may be a tool YC itself built to understand how applicants use AI, not some third-party data-harvesting operation. That’s arguably less scary from a privacy standpoint. But it doesn’t change the fundamental problem. It might even make it worse.

Because if YC built the tool themselves, they should have known exactly what would happen next.

When you tell people what you’re measuring, smart people stop doing the thing you’re measuring and start doing the thing that scores well.

This is Goodhart’s Law with a Silicon Valley budget. And it’s already producing a pathology that one commenter identified with surgical precision: gstack.

Gstack is what happens when you believe too hard in secret metrics. People discover the correlations the system rewards, then chase those correlations until they don’t mean anything anymore. You think you’re optimizing for founder quality. You’re actually optimizing for the appearance of founder quality. And the gap between those two things is where entire ecosystems go to die.

Think about the incentive structure this creates. You’re a ambitious 24-year-old hacker. You could spend the next three months building something genuinely novel — something that might fail, something that might teach you something real. Or you could spend three weeks reverse-engineering Paxel’s scoring logic, curating your codebase to hit the right patterns, and gaming your way into a $500K check and the most powerful alumni network in tech.

Which do you choose?

Be honest.

The system doesn’t reward the best founders. It rewards the ones who optimize for the system. Those are not the same people, and pretending they are is how you get an industry full of metric-chasing zombies.

YC isn’t stupid. They know this risk exists. But here’s the uncomfortable truth: at the scale of 100,000+ applicants, you NEED automation. You need AI to filter. No human team can read that many applications. So you deploy the tool, you accept the trade-off, and you hope the gamers stay a small enough minority that the signal still outweighs the noise.

But that hope is a bet, not a strategy. And every time a post like this goes viral, the bet gets worse. Because now MORE people know the game exists. More people will optimize for the evaluator instead of the market. More founders will spend their cognitive energy on meta-cognition — thinking about what YC’s AI wants to see — instead of building things people actually want.

This is the arms race nobody’s talking about. Not the AI arms race between nations. The one between gatekeepers and applicants, where every tool designed to find merit creates a new incentive to fake it.

And here’s the darkest part: it’s not even clear the hackers are wrong.

If YC’s system can be gamed, maybe the system was never measuring what it claimed to measure in the first place. Maybe there’s no reliable signal in a coding-agent-generated report about your git history. Maybe the whole exercise is theater — expensive, scalable, AI-powered theater that makes everyone feel like a rigorous process is happening when really it’s just pattern-matching on noise.

Every metric becomes a target. Every target becomes a game. And every game, eventually, gets hacked. The only question is whether you notice before the players take over the casino.

So where does this leave us?

If you’re a founder: you now face a choice that didn’t exist five years ago. Do you build something real, or do you learn to play the meta-game? The honest answer is that the system rewards both, but it punishes neither clearly enough to make the choice obvious. That ambiguity is the real danger.

If you’re an investor: your evaluation tools are creating the very behavior they’re supposed to detect. Your AI scoring system isn’t a window into founder quality. It’s a mirror reflecting whatever patterns you trained it to reward. Look closely at what’s staring back.

If you’re YC: someone just told 100,000 founders that your front door has a trick lock. You can change the lock. You can pretend you don’t hear the knocking. But you can’t un-ring this bell.

The meritocracy was always partly a myth. But at least it used to be a myth that pointed in the right direction — build something people want, and you’ll get rewarded. Now the myth points at the evaluator. Build something the machine wants to see, and you’ll get rewarded.

Those are different gods. And only one of them cares about whether your product actually works.