# 48 AI agents found a Maya site overnight. Archaeologists had already mapped it.

<!--
title: "48 AI agents found a Maya site overnight. Archaeologists had already mapped it. · Cork"
description: "Pietro Miotti's 48 AI agents went through 1,200 km² of NASA lidar overnight and flagged a Maya site Mexico had no record of. A 2025 paper had already mapped it. Agents are good at a week of work while you sleep; knowing what they didn't check is the hard part."
canonical: "https://corkcli.com/blog/ai-agents-maya-site-already-mapped"
last_updated: "2026-11-16"
published: "2026-11-16"
author: "Kevin Vera"
-->

Pietro Miotti's 48 AI agents went through 1,200 km² of NASA lidar overnight and flagged a Maya site Mexico had no record of. A 2025 paper had already mapped it. Agents are good at a week of work while you sleep; knowing what they didn't check is the hard part.

By Kevin Vera · November 16, 2026 · [News](https://corkcli.com/blog/c/news) · 5 min read

In October 2026 Pietro Miotti, a Google researcher and, as he says up front, not an archaeologist, set 48 AI agents loose on public NASA lidar of the Maya lowlands and went to bed. By morning they had flagged a hilltop site that Mexico's National Institute of Anthropology and History (INAH) had no record of. A few weeks later, archaeologists pointed out that it was already in a peer-reviewed paper. Miotti told The Verge the project was a personal one, done without any Google compute or resources. What it shows is how good agents have become at doing a week of work overnight, and how hard it is to see what they skipped.

## What happened

Miotti tells it as a diary on his blog:

- **18 October.** Travelling in Peru, he wondered what Hiram Bingham, the explorer who "discovered" Machu Picchu in 1911, would have done with today's tools. The agents he asked (Gemini, ChatGPT, Claude and Grok) all said lidar, which fires laser pulses through gaps in the canopy and maps the ground underneath. Peru isn't well covered by public lidar; the Maya region is. "That night, I asked the agents to analyze it and went to bed."
- **19 October.** The agents had turned the lidar strips, "about 1,200 km² of jungle", into images, and "48 agents went through them one by one", looking for pyramids, platforms and plazas while filtering out natural hills, sinkholes and roads. "There was no special model or classifier involved: the agents literally looked at the pictures, following a precise guide." He asked them to check which villages hadn't been discovered yet: "Search both the literature and the archives." A few hours later: "One major village doesn't seem to have been registered." On a hilltop at the edge of a cliff: a levelled plaza, "an oval pyramid at least 8–12 m tall", a raised courtyard and a hollow that might have been a reservoir.
- **20 October.** The agents checked the raw point cloud, to make sure the shapes were on the ground and not made by the trees. The report, translated into Spanish, went to INAH.
- **21 October.** INAH replied that it had checked its records and found no registered archaeological site at that location. "It took less than 72 hours to go from a random thought while traveling to an official reply from the Mexican government."
- **11 November.** After the story spread, archaeologist Sarah Parcak pointed out on Bluesky that Tulane's Francisco Estrada-Belli had already published the site. Miotti updated his post the same day: "The agents missed it while going through the literature, so they rediscovered it rather than discovered it." The site is in Estrada-Belli, Canuto, Šprajc and Fernandez-Diaz, a 2025 lidar survey in the Journal of Archaeological Science: Reports.

## What the agents missed, and why

The image work held up. Elizabeth Graham, a professor of Mesoamerican archaeology at UCL, credited Miotti with doing a decent job of mapping the mounds from lidar, noting that "the interpretation of lidar data can be tricky."

The miss was in the step after: checking whether anyone had found it before. Miotti explained it to The Verge: "My agents and I missed it. Archaeologists rightly don't publish the coordinates of sites, so checking whether a site is already known means comparing images against figures in papers, not searching a database. That's much harder for agents, and it's something these tools need to get better at."

So the agents searched what was searchable, came back clean, and the clean answer looked the same as one from a search that could have found the paper. INAH's register didn't settle it either. Ivan Šprajc, one of the paper's authors, told The Verge that checking with INAH is "not sufficient": "The fact that INAH has no official knowledge of a particular cluster of archaeological features does not necessarily mean that it has not been spotted or even explored by archaeologists." He also said he was "a little bit concerned about being flooded" by messages from AI "discoverers".

## It isn't really about archaeology

Miotti says so himself: "So far, most agentic AI stories have been about coding, math and physics, but it's much more general than that." He meant what agents can do. It holds for what they miss, too. An agent reports what it found, and rarely what it couldn't check. A coding agent that says "there's no helper for this" searched for the names it guessed. One that says "the tests pass" ran some tests. Sending 48 agents at a job overnight makes the result arrive faster, and makes the question of what they actually did in the meantime bigger.

A few habits make overnight agent work reviewable:

- **Ask for what was checked, not only what was found.** "Searched the register and these three databases; could not compare against figures in papers" tells a reviewer exactly where to look.
- **Keep finding and verifying separate.** The agent that found something is the wrong one to confirm it is new.
- **Ask someone who knows the field before announcing.** Here, one Bluesky post settled it.
- **Keep a record of what each agent did.** When the morning result is surprising, you need the steps that produced it, not only the report.

Cork works on that last one for coding agents. It keeps a live record of what every agent on a repository changed, file by file and function by function, across Claude Code, Codex and Cursor, and its `recap` tool says what happened and what it cost. It doesn't check whether a paper exists. It does make "what did the agents do last night?" a question with an answer. (More on [seeing what all your AI coding agents are doing at once](https://corkcli.com/blog/see-what-all-ai-coding-agents-are-doing).)

## Sources

Checked 16 November 2026:

- Pietro Miotti, "My AI Agents (Re)discovered a Maya Site in 72 Hours", with its 11 November update: https://pietromiotti.com/latest/maya-site/
- Robert Hart, "An AI agent swarm may have unearthed an unmapped Maya site.", The Verge, 11 November 2026: https://www.theverge.com/ai-artificial-intelligence/1031498/an-ai-agent-swarm-may-have-unearthed-an-unmapped-maya-site
- Robert Hart, "No, AI agents did not discover a previously unknown Maya site", The Verge, 12 November 2026: https://www.theverge.com/ai-artificial-intelligence/1032125/ai-agents-maya-site-archeology-discovery
- "AI agents found a Maya site. Archaeologists found it first", AI Scoop News, 12 November 2026, for the 2025 paper's title and authors (Estrada-Belli, Canuto, Šprajc and Fernandez-Diaz, "New regional-scale Classic Maya population density estimates and settlement distribution models through airborne lidar scanning", Journal of Archaeological Science: Reports, 2025): https://aiscoopnews.com/posts/ai-agents-maya-site-already-discovered

## Frequently asked questions

### Did AI agents discover a new Maya site?

No. Pietro Miotti's agents found a site that wasn't in INAH's register, but it had already been mapped in a 2025 lidar survey by Estrada-Belli and colleagues. Miotti updated his post to say the agents "rediscovered" it.

### How did 48 AI agents analyze the lidar?

According to Miotti, they turned about 1,200 km² of NASA lidar strips into images and went through them one by one, following a written guide, filtering out natural hills, sinkholes and roads. No special model or classifier was involved. They then checked the raw point cloud to confirm the shapes were on the ground.

### Why did the agents miss the earlier paper?

Archaeologists don't publish site coordinates, so finding earlier work means comparing images with figures in papers, not searching a database. Miotti told The Verge that is much harder for agents.

## Related reading

- [Cork vs Dust: company agents, or coding agents on one repo](https://corkcli.com/blog/cork-vs-dust): Cork vs Dust: Dust is a platform for building agents on a company's knowledge and tools, for every team. Cork coordinates the coding agents a team already runs (Claude Code, Codex, Cursor) so they don't overwrite each other on one codebase. Where each wins, and how they work together.
- [One goal, then the agents did the rest: what the FBI's ARTEX warning means for your team](https://corkcli.com/blog/fbi-artex-warning-ai-agent-intrusions): The FBI's FLASH on ARTEX describes AI agents that take a goal in plain language and carry an intrusion from scouting to stolen credentials, some reaching privileged access in under three hours. When an attack finishes that fast, an alert is too late; the controls that work are the ones already in the path.
- [The attacker kept office hours, the agent did the work: inside CloudSEK's Strix report](https://corkcli.com/blog/strix-ai-agent-cloudsek-silent-lynx): CloudSEK found an attacker's server open on the internet, and on it an AI agent named Strix: tasked over Telegram, given a persona file and pre-approved destructive commands, with a tunnel into Tajikistan's national railway network for nearly two months. Its building blocks are the ones teams use for their own agents.
- [Agent merge conflict: when parallel AI agents' branches will not merge](https://corkcli.com/blog/agent-merge-conflict): An agent merge conflict is a git merge conflict between branches written by AI coding agents running in parallel, usually found only when the work is combined.

[All articles](https://corkcli.com/blog.md) · [Cork in Markdown](https://corkcli.com/index.md) · [llms.txt](https://corkcli.com/llms.txt)

This page in HTML: https://corkcli.com/blog/ai-agents-maya-site-already-mapped
