According to co-founder Grace Li, her company was born just weeks before graduation in 2025, when a group of college friends set out to refine their AI game engine. While the models could produce functional games, none managed to capture the element of fun, leading to the pivotal question: how do you determine a game’s fun factor?
Realizing that human judgment was irreplaceable, they began brainstorming methods to gather genuine human feedback on a larger scale. This endeavor evolved into Design Arena, an AI tool currently utilized by 5.3 million users globally. It emerged that many AI firms were in search of scalable user feedback and were prepared to invest in solutions.
“It filled a crucial gap for numerous models aiming to enhance design capabilities,” Li explains. “Shortly thereafter, we secured our first significant agreement with a frontier lab, and the rest followed.”
On Monday, the company behind Design Arena, named Intelligence, revealed it had successfully closed a $7.9 million seed funding round led by Index Ventures, with contributions from other investors such as Conviction, A*, and Valkyrie.
For casual users, Design Arena operates similarly to a complex model router. Users interact through a ChatGPT-like prompt window, selecting from various formats such as websites and images. Upon submitting their request along with preferred styles, users are presented with multiple options to compare and rank from best to worst.
While this is a helpful service, the true strength of the platform is on the enterprise level, where models can leverage it as an endless source of immediate feedback for their media creation processes. Users typically don’t have a preference for which models they evaluate — they prioritize receiving the highest quality output — making their rankings essential for understanding user desires.
According to Li, for frontier labs, this feedback service holds significant value, with the company currently generating $60 million in annual recurring revenue, establishing itself as an indispensable source of human evaluation data within the AI sector.
Importantly, users are required to log in to access their results, allowing Intelligence to monitor how preferences evolve across different regions and timeframes. Li remarks that web aesthetics in Asia often lean towards a more maximalist design approach. This data serves as a crucial addition to automated benchmarks that, while expansive, can easily be manipulated, as seen in the recent breach at Hugging Face.
However, relying solely on crowdsourced human feedback does not guarantee market success. For instance, Yupp closed its doors less than a year after launching despite raising $33 million from a prominent crypto investor and claiming over 1.3 million users, failing to establish a sustainable business model.
Nevertheless, other startups focused on human evaluation appear to be flourishing. For example, LM Arena, which employs a comparable approach for text responses, secured $150 million in Series A funding just four months after launching its paid service.
