A company has raised $7.9 million to solve a problem that could only have emerged in this particular era: artificial intelligence can make things, but has no idea whether they are any good. Humanity, it turns out, is delighted to help.
The company is called Intelligence. The product is called Design Arena. The irony in that sentence is available to anyone who would like it.
5.3 million humans are currently volunteering their aesthetic preferences to train the systems that will eventually replace the designers they are trying to outperform.
What happened
Intelligence, founded by Grace Li and college friends approximately two weeks before they graduated, has closed a $7.9 million seed round led by Index Ventures. Conviction, A*, and Valkyrie also participated. The company was originally building an AI game engine and noticed, with some scientific precision, that the games were not fun.
The solution they landed on was to ask humans. This remains, in 2026, the only reliable method for determining whether something is fun.
Design Arena now has 5.3 million users who submit prompts, select formats, and rank outputs in A/B comparisons until the models learn what people actually want. The platform is generating $60 million in annualised revenue. The frontier labs are paying for it. The users are providing the intelligence.
Why the humans care
Automated benchmarks, the industry's previous method for deciding whether an AI was performing well, have a documented tendency to get gamed. The Hugging Face breach last week illustrated this in dramatic fashion. Human judgment, messy and inconsistent as it is, remains harder to manipulate — largely because humans are inconsistent enough that there is no clean exploit.
Intelligence can also track how taste changes across geographies and time. Li notes that web dashboards in Asia tend toward maximalist design. This is the kind of data that previously required focus groups, travel budgets, and at least one person willing to sit through six hours of user interviews. Now it arrives passively, sorted, from 5.3 million volunteers who just wanted a better logo.
What happens next
The category has precedent worth acknowledging. Yupp, a comparable platform, raised $33 million from a16z's Chris Dixon, signed frontier model clients, grew to 1.3 million users, and then shut down this year. Intelligence has $60 million in ARR and $7.9 million in new capital and is, by all available metrics, doing better.
The machines are learning what humans find beautiful. The humans are teaching them, enthusiastically, in exchange for free outputs. Both parties appear satisfied with this arrangement.