Google SWE deploys 48 AI agents at NASA Lidar, uncovers unknown Maya pyramid in 72 hours
agents google
| Source: HN | Original article
A Google software engineer deployed 48 AI agents to analyze NASA Lidar data, uncovering an unmapped Maya pyramid within 72 hours.
An independent software engineer has shown that a coordinated swarm of large‑language‑model agents can turn raw remote‑sensing data into a publishable archaeological discovery in days. By orchestrating 48 frontier AI agents—including Gemini, ChatGPT, Claude and Grok—the developer fed 1,200 km² of NASA LiDAR point‑cloud data covering dense Mexican jungle into an automated pipeline. Within 72 hours the agents had converted the strips into visual mosaics, filtered out natural terrain, and flagged a hilltop that matched the layout of a Classic Period Maya civic centre: a leveled plaza, an oval pyramid and surrounding platforms that do not appear on any existing map.
The breakthrough demonstrates that large‑scale, unsupervised multi‑agent workflows can extract high‑value signals from massive geospatial datasets without continuous human supervision. It validates Google’s SWE (Software Engineering) framework for deploying dozens of LLM instances in parallel, and suggests a new model for remote‑sensing archaeology where AI does the first pass of pattern recognition, leaving specialists to verify and interpret the findings. The rapid turnaround—from concept to an official response from the Mexican government—highlights how AI could compress years of field scouting into a matter of hours.
The next steps will likely involve on‑the‑ground verification of the site, peer‑review of the methodology, and broader testing on other heritage‑rich regions. Researchers will watch for follow‑up studies that integrate additional satellite modalities, assess the false‑positive rate of such swarms, and explore how governments and cultural‑heritage bodies will regulate AI‑driven discoveries. If the approach scales, it could reshape data‑intensive fields from archaeology to environmental monitoring.
Sources
Back to AIPULSEN