Google Engineer Uses AI Agents to Discover Unmapped Maya Pyramid
Forty-eight AI agents analyzed NASA lidar data to identify an ancient site in Mexico.
1AI-Powered Archaeological Discovery
A Google software engineer utilized 48 AI agents to analyze NASA lidar data, successfully identifying a previously unmapped Maya site in Mexico. The process, which spanned only 72 hours, involved scanning 1,200 square kilometers of jungle terrain. The discovery includes a leveled plaza, an 8–12 meter tall pyramid, and a potential water reservoir.
2Lidar Data Processing Mechanism
Lidar technology fires laser pulses through dense jungle canopies to map ground structures hidden from satellite imagery. The agents converted lidar strips into visual images, systematically filtering out natural hills, sinkholes, and roads. By following a precise visual guide, the agents identified straight-edged structures that typically indicate human-made architectural features rather than nature.
3Agentic Workflow and Verification
The project employed a multi-agent system including Gemini, ChatGPT, Claude, and Grok to cross-reference findings against existing literature and archives. After identifying the site, the agents drafted a formal report in Spanish and provided contact instructions for Mexico's National Institute of Anthropology and History, known as INAH, to verify the site's status.
4Official Response from INAH
Following the submission of the report, INAH officials confirmed that no archaeological site was previously registered at the specified coordinates. While the discovery remains unofficial pending a physical field inspection, the initial analysis suggests a Classic period civic center. The site likely served as a hub for ceremonies, markets, and local governance.
5Historical Context of Maya Sites
The site likely dates to the Classic period, roughly AD 250–900, featuring a raised courtyard and platforms that likely served as residential areas. The presence of a flat-bottomed hollow suggests an aguada, a critical rainwater reservoir used by the Maya to survive the dry season, indicating a sophisticated level of urban planning.
6Broader Scientific Research Trends
This project builds upon recent advancements in remote sensing, such as the 2024 discovery of an entire Maya city and studies estimating population density in Maya lowlands. Researchers are increasingly reusing environmental lidar data for archaeological purposes, demonstrating that existing datasets hold vast, untapped potential for identifying ancient human settlements across the globe.
7The Role of Naive Outsiders
The engineer attributes the success to a naive approach, bypassing traditional academic bottlenecks by applying computational power to an old problem. By treating the discovery as a general task rather than a specialized archaeological study, the project highlights how non-experts can leverage AI agents to accelerate breakthroughs in diverse scientific and historical fields.
8Future of Agentic Science
This experiment suggests that the future of scientific discovery will be limited only by human curiosity and available compute power. As agentic AI moves beyond coding and physics, the potential to map the entire planet using existing lidar data becomes a tangible goal, marking a significant shift in how we conduct large-scale field research.