Grace Lee's Intelligence raises seed from Index to build a 5.6M-user design arena that trains AI on human creative preferences

Intelligence raises seed from Index Ventures with ~$60M in revenue already on the books

Grace Lee's company Intelligence, the team behind Design Arena, just came out of stealth. The timing is notable: they closed a seed round led by Index Ventures after already reaching approximately $60M in revenue and 5.6 million users across 192 countries.

The business started as a game engine, then pivoted when a friend posted a link on Reddit without telling them and 3,000 users showed up overnight. That was the signal. Design Arena lets users generate multiple versions of a design simultaneously, collect feedback on which resonates, and use that preference signal to improve frontier AI models. The form factor mirrors how designers actually work with clients — show options, read reactions, iterate.

“We just one of our friends linked it on Reddit. He didn't even tell us about it, and it got, like, 3,000 users overnight. We're like, well, that's more than the game engine has ever gotten. And since scaled that to, oh, man, 5,600,000 people across 192 countries. On the Frontier Lab side, we pretty much work with all the major players to help improve their models and hard to verify domains.”

How the preference flywheel works

The product sidesteps the cold start problem because it doesn't ask users to do extra work. Showing multiple design versions is what a designer does anyway, so the feedback is a natural byproduct of using the tool. Lee frames the company's core job as acting like a PM: figure out what "better" means for each specific user, track the right signal, and feed that back to the models. A user building a portfolio optimizes for aesthetics; a user building a conversion-focused landing page optimizes for dollars at the last screen. The label applied to the preference data changes accordingly.

Lee is explicit that improving models doesn't threaten the business. Better models make Design Arena more useful to more people, which generates more preference data, which improves the models further.

The bigger ambition

Lee's longer-term thesis is an intelligence marketplace, where Intelligence becomes the layer that matches suppliers of AI models with people seeking intelligence outputs — a positioning she compares to PageRank's role in making the internet accessible. The analogy may be ambitious, but the underlying logic is that preference data at scale is the defensible asset, not any single model relationship.

On creative superintelligence, Lee's framing is worth noting: she defines it not as AI generating original ideas, but as collapsing the distance between an idea and its existence in the world. A person who can imagine a new pen design shouldn't need CAD skills, manufacturing contacts, or a supply chain. That framing keeps the human as the originator and the model as the instantiation layer — at least for now.

Gaming Arena

Lee confirms the company is building a gaming equivalent of Design Arena. Game development is an especially hard domain to verify because "fun" is difficult to measure without real players, which is exactly why the arena format applies. She describes it as putting a game on the equivalent of the internet and seeing if it gets picked up — with platforms like Steam and Instagram's algorithm as the real-world analogs for what the arena is trying to replicate.

At ~$60M in revenue on a seed round, the ratio of revenue to funding is unusual enough that the company's stealth until now is the more surprising fact.