AI Agents Reanalyze Public Lidar and Rediscover a Maya Site

A developer with no archaeology background pointed a swarm of AI agents at public lidar data overnight and woke up to what looked like an unrecorded Maya settlement under the Mexican jungle. The twist, added in an update later, is that specialists had already mapped the site. The story is still worth reading, because it shows how far a determined outsider can now get with off-the-shelf tools and open data.

From a Machu Picchu thought to a Maya hilltop

The idea started during a trip to Peru, while reading about Hiram Bingham and his 1911 arrival at Machu Picchu. Several chat assistants agreed on the right tool for finding ruins hidden by forest: lidar. Satellite photos see only treetops, but lidar fires laser pulses, some of which slip through gaps in the canopy and bounce off the ground, producing a map of the terrain underneath.

Peru has little public coverage, and commissioning a survey would have meant permits and cost. The Maya lowlands are far better served, including airborne lidar originally collected for environmental research and later reused by archaeologists. That pattern is well established: in 2024 researchers reported a whole Maya city in Mexico found in existing lidar, and other teams have mapped settlements across the region the same way.

How the agents worked

There was no custom model or trained classifier. About 1,200 square kilometres of lidar were converted into images, and 48 agents reviewed them one by one, following a detailed written guide. Their job was to spot pyramids, platforms and plazas while screening out natural hills, sinkholes and roads.

The overnight run turned up many village-like clusters. A second pass asked the agents to check the literature and archives for any that were not yet recorded. One large hilltop site on a cliff edge seemed to be missing. It showed straight edges and sharp corners, a levelled plaza, an oval pyramid estimated at 8 to 12 metres tall, an attached raised courtyard and a flat-bottomed hollow that might have held rainwater. Checking the raw point cloud confirmed the shapes were real ground features rather than artefacts of vegetation.

The author then wrote a report with the agents, translated it into Spanish and sent it to Mexico's National Institute of Anthropology and History. INAH replied that it had no registered site at that location.

Rediscovered, not discovered

The update changes the headline but not the lesson. The site had been published in a 2025 study of Maya lowland settlements. The agents missed it during the literature check, so they rediscovered it. It simply was not yet in INAH's register, which explains the official reply.

That gap matters. Two things can be true at once: the method reached the same place as professional researchers, and the literature-search step was the weak link. Agents that confidently report "nothing found" in prior work can be wrong, and anyone building similar pipelines should treat negative search results as claims to verify, not facts.

Why developers should care

The archaeology is almost beside the point. The notable part is the shape of the workflow:

• Public, openly licensed data, here airborne lidar gathered for other purposes.
• A swarm of general-purpose agents doing repetitive visual review that once needed trained staff.
• A human who set the goal, supplied judgment and handled the real-world follow-up with an institution.

The timeline from idea to an official government reply was under 72 hours. Cheap parallel review of large datasets is now within reach of individuals, and the bottleneck shifts to verification, domain knowledge and knowing whom to report to.

What to watch

Expect more outsiders to run this kind of survey on other regions and datasets, and expect false positives and rediscoveries to be common. The useful next steps are better automated checks against existing publications, clearer routes for submitting candidate sites to heritage bodies, and closer collaboration with the archaeologists who hold unpublished results. The author has said the site is now in the hands of INAH and the researchers who mapped it first, and promised an update if anything comes of it.