TLDRocket
Sign in

DesignArena creators raise $7.9 million to bring taste to AI models

TechCrunch AI Russell Brandom

A startup that turned game design flops into an AI feedback business just raised $7.9M. Turns out AI labs will pay big for human taste data — this one's already at $60M ARR.

Grace Li and her college friends set out to build an AI game engine. What they actually built, almost by accident, is a company that sells human taste as a service. The games their engine produced technically worked, but nobody enjoyed playing them, and that gap between "functional" and "fun" turned into the whole business.

Their fix was to stop guessing and start asking people directly. DesignArena shows users A-versus-B comparisons — websites, images, dashboards, a dozen other visual formats — and has them rank outputs from best to worst. Individual users mostly don't care which model made what; they just want the better result. That indifference is exactly what makes the rankings valuable to the AI labs paying for them, because it strips out brand bias and leaves pure preference data behind.

The pivot happened fast. Li says a frontier lab signed on within a week of the idea taking shape, and DesignArena now claims 5.3 million users and $60 million in annual recurring revenue. Because people have to log in to see their results, the company can also track how taste shifts by region — Li points to Asia's dashboards trending more maximalist as one example — which is the kind of granular signal automated benchmarks simply can't produce.

That benchmark gap matters more than it might sound. Standard leaderboards can be gamed, and last week's Hugging Face breach showed just how exposed those systems are. Human ranking, done at scale, is harder to manipulate in the same way, which is a big part of why labs are willing to write checks for it.

But betting on crowdsourced feedback isn't a guaranteed win. Yupp raised $33 million from Chris Dixon, landed frontier-model clients, hit 1.3 million users, and still folded within a year. LM Arena, running a similar playbook for text, just raised $150 million four months after launching its paid product. Same basic idea, wildly different outcomes — which suggests execution and timing matter here as much as the concept itself.

The $7.9 million seed, led by Index Ventures with Conviction, A*, and Valkyrie joining in, gives Intelligence room to scale a business that started as a side quest to fix boring AI-generated games.

My take

This is the most sensible AI-adjacent business I've seen this quarter: sell the thing benchmarks can't fake, which is actual human preference. But Yupp's collapse is the tell nobody should ignore — having frontier-lab customers and a million users isn't a moat if the underlying data commodifies fast, and I'd bet at least one of DesignArena's current clients is quietly building an in-house version right now.

Read more about this at: TechCrunch AI

Related stories

The daily briefing

Every AI story that matters, in your inbox by 8am.

TLDRocket reads 60+ sources, removes duplicate coverage, and summarises the day in two minutes. Follow companies and topics for alerts, or get the briefing in Slack. Free, no spam, unsubscribe anytime.