Human API has emerged from stealth with a platform that lets AI agents directly assign tasks requiring human judgment to people. Developed by the Eclipse team, which built a high-speed Ethereum layer-2, the startup addresses AI's "last mile" problem where physical-world interaction is essential. It begins by providing audio datasets capturing linguistic nuance to AI labs.

Sydney Huang, founder of HumanAPI and CEO of Eclipse, explains that agents fail at tasks needing real-world endpoints, like package delivery. The initial business model licenses studio-quality speech recordings, having already produced data for a frontier AI lab. This tackles the need for specific human input to distinguish regional accents or pronunciations.

The company has raised $65 million from investors including Placeholder and Polychain Capital. Plans include expanding beyond audio into computer-usage data and logistics tasks. The ultimate goal is a simple API call for agents to source human execution, creating compensated micro-tasks while giving AI developers structured human input.

This approach could redefine how AI systems handle real-world tasks while creating new income streams for human workers, moving beyond pure automation where human nuance remains irreplaceable.