The Graveyard of Correct Ideas
5,700 dead YC startups. Most of them had real ideas. That's the uncomfortable part.
Victor Asemota asked Claude to analyse startups.rip — a directory of over 5,700 failed YC startups with post-mortems and rebuild plans. What came back wasn’t a list of cautionary tales about bad ideas. It was a list of correct ideas that died anyway.
That’s the uncomfortable part.
Timing is brutal and non-negotiable
Sam Altman built Loopt in 2008. Location sharing between friends, friend-finding on a map, real-time check-ins. You know this product — it’s Google Maps, Life360, and Find My Friends. The idea was right. The moment was wrong. Smartphone infrastructure, social norms around location sharing, carrier support — none of it was ready. Loopt died. The idea is now worth billions in other hands.
Airo Health built a stress-monitoring wearable and launched in August 2016. Apple Watch Series 2 launched in September 2016. Fitbit added stress tracking the same month. Game over.
These weren’t bad founders or bad ideas. They were correct ideas with catastrophic timing. The market catches up — but rarely on your runway.
Find a sharp edge. Don’t blunt it.
Posterous launched with an elegant premise, grew 30% month-over-month, and hit 15 million monthly visitors. Then it abandoned its winning position to fight a social media war it was never built for. Feature accumulation eroded the product’s clarity. Monthly visitors collapsed from 15 million to 1.33 million before Twitter acquired the team — not the company.
This pattern repeats across the graveyard constantly. Find a sharp edge, then blunt it chasing a bigger vision. The acqui-hire is the startup world’s equivalent of a graceful funeral — the company dies but the founders land softly, which obscures how common actual failure really is.
The AI wrapper problem
The W22–S23 YC cohort is instructive: Flike, Parabolic, Fabius, CoffeeAI, Dialect, Demo Gorilla — almost all AI-wrapper companies doing one thing. AI sales emails. AI customer support. AI outreach. When ChatGPT can do 80% of what your product does for $20/month, your margin for differentiation becomes razor-thin.
These companies weren’t solving hard problems. They were building thin layers on top of foundational models that rapidly caught up with them. A feature is not a company.
Don’t compete with your supply chain
42floors needed broker data to populate listings. Then they became brokers themselves. Brokers stopped sharing data. Company died.
Tutorspree took 50% of tutor fees — giving both sides a massive incentive to bypass the platform the moment they’d connected. The business model had a structural flaw baked in from day one.
The meta-pattern
What startups.rip reveals most clearly is that most failures weren’t execution problems. They were structural problems — market timing, business model mismatch, building on quicksand.
The graveyard is full of real ideas. Some of them are now worth billions in other hands. The gap between insight and execution, between timing and readiness, between building something people want and building something that sustains itself financially — that’s where it fills up.
You don’t just need a good idea. You need the right structure, the right moment, and a model that doesn’t cannibilise itself.
Ideas don’t die. Startups do.
Analysis sourced from Victor Asemota’s thread on startups.rip. Credit and gratitude to him for doing the heavy lifting.