Google SWE points 48 AI agents at NASA Lidar, finds unmapped Maya pyramid in 72h

A software engineer used 48 AI agents to inspect public lidar data covering about 1,200 square kilometres of Mexican jungle, hoping to find Maya remains hidden beneath the trees. Lidar maps the ground through gaps in forest canopy, making it possible to spot shapes that ordinary satellite images conceal. The agents converted the data into images and visually examined them using a guide to distinguish possible buildings from natural landforms, roads and sinkholes; no specialized detection model was used. Among the features they flagged was a hilltop complex with a level plaza, platforms and an oval pyramid estimated to be 8-12 metres tall. Checks of the raw lidar points suggested these shapes belonged to the ground, not the vegetation.

After reviewing archaeological literature and archives, the engineer reported the site to Mexico’s National Institute of Anthropology and History. The agency replied that it had no registered archaeological site at the specified location, but the find is not yet officially confirmed: authorized fieldwork would be needed, and it is up to local authorities whether to investigate. The proposed interpretation is a small Maya civic centre from roughly AD 250-900, with a temple, gathering plaza, possible elite residence and perhaps a rainwater reservoir; these identifications remain speculative. The account presents the discovery as a demonstration of how AI agents can apply patient visual analysis to scientific problems, while arguing that lidar and public data could help researchers find more sites.