A software engineer's claim that his fleet of AI agents found a previously unregistered Maya settlement in Mexico went viral this week, drawing nearly a thousand upvotes on Hacker News — before acquiring a twist that may be more interesting than the original discovery. The site, it turns out, had already been mapped by professional archaeologists. The agents simply didn't know, and neither, initially, did Mexico's national archaeology institute.
The story, documented in a first-person blog post by Pietro Miotti and last updated November 11, began on October 18 while he was traveling in Peru. Musing on Hiram Bingham's 1911 "discovery" of Machu Picchu, he asked a group of AI agents — Gemini, ChatGPT, Claude and Grok — what a modern explorer should do. Their unanimous answer: lidar, the laser-mapping technique that fires pulses through forest canopy to reveal the ground beneath. For more stories at the intersection of AI and discovery, see our AI news coverage.
48 agents, 1,200 square kilometers, one night
Peru's lidar coverage proved sparse, so the project pivoted to the Maya region, where public lidar data is abundant. Miotti asked the agents to analyze roughly 1,200 square kilometers of jungle lidar strips overnight. Their method was strikingly low-tech: the strips were converted into images, and 48 agents examined the tiles one by one "like very patient archaeologists," following a written guide to distinguish pyramids, platforms and plazas from natural hills and sinkholes. No custom model or classifier was involved — the agents literally looked at the pictures.
By morning, the agents had flagged numerous rectangular structures, and a literature-checking pass concluded that one hilltop settlement did not appear in any register. Zooming into the lidar revealed a leveled plaza, an oval pyramid at least 8–12 meters tall, a raised courtyard, and a depression that might have been a rainwater reservoir — plus three smaller platforms likely used as houses. The agents then verified the features in the raw point cloud, confirming that the ground itself — under 10–17 meters of forest — traced those shapes.
INAH says: not in our register
On October 20, the agents drafted a formal report, translated it into Spanish, and Miotti sent it to INAH, Mexico's National Institute of Anthropology and History. The reply came the next day: INAH checked its records and found no registered archaeological site at the coordinates provided. "At that point I thought nobody had put it on a map before," Miotti wrote. "I was wrong."
In an update published November 11, he acknowledged that researchers had in fact mapped the same site in a 2025 study led by Francisco Estrada-Belli and colleagues covering settlements across the Maya lowlands. The agents' literature review had missed the paper, and the site had never entered INAH's official register — hence the "rediscovery." Miotti takes the correction in stride: "an outsider with agents ended up in a similar place to the experts. And I don't think this is really about archaeology."
The episode lands in a fast-moving research area. In 2024, archaeologists including Auld-Thomas and colleagues reported an entire previously unknown Maya city found in public lidar data, and separate work published this year has identified what may be ancient Maya marketplaces. Miotti's experiment suggests the screening step of that pipeline — visually scanning vast lidar coverage for candidate structures — can now be delegated to commodity AI agents working overnight.
Promise, with a built-in caveat
The Maya story carries a lesson its author states plainly: agentic AI is breaking out of coding, math and physics into every scientific field, and the bottleneck is shifting to "curiosity and imagination (and compute)." At the same time, the rediscovery caveat is the story's own footnote on AI limitations — the agents' literature search failed precisely where expert knowledge mattered, and the find only became real once it was placed in the hands of INAH and the archaeologists who mapped it first.
Miotti, who is not an archaeologist, is careful about the site's interpretation. He describes it, based on agent-assisted speculation, as a small Classic-period civic center — roughly AD 250–900 — where a plaza hosted gatherings, a temple pyramid anchored ceremonies, and a local elite lived beside the raised courtyard. Verification, he notes, belongs to the professionals: "If anything comes out of it, I'll post an update."
The viral reaction — the post drew 955 points on Hacker News within a day — suggests the public appetite for AI-assisted discovery extends well beyond software. Whether the practice becomes standard archaeology or remains an outsider's stunt may depend on the next step Miotti himself suggests, only half-jokingly: lidar the whole planet and let the agents find the rest.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
