Google researcher Pietro Miotti set 48 AI agents to comb NASA lidar data for signs of ancient sites after wondering, during a trip to Peru, how explorers might use modern technology. With public lidar coverage there limited, he turned to the Maya region. An early account said the agents flagged a possible site in Mexico, while stressing that Miotti was not an archaeologist and that fieldwork was still needed to verify it.¹

Miotti then contacted Mexico’s National Institute of Anthropology and History (INAH) to ask whether the location was already known, including through unpublished work. He said the agency had no record of a registered site there. He described the project to The Verge as a personal side project, without Google computing resources, and said AI agents helped translate his correspondence into Spanish.²

But the apparent discovery did not hold up as new. A later report said the site had already been identified and published in an academic journal the previous year. It also noted that INAH’s reported lack of a registered record did not establish that the location was unknown: research may be unpublished, and local communities can know of archaeological sites before they enter formal records.²

The two accounts captured both the promise and the limits of AI-assisted research. The agents may have helped surface a location worth examining, but identifying it as a discovery required context beyond the lidar scan. The correction also sharpened a broader concern among archaeologists: powerful tools can help outsiders search specialist fields, while lacking the expertise to tell a promising lead from work already done.