The author deployed multiple LLM agents (Gemini, ChatGPT, Claude, Grok) to analyze approximately 1,200 km² of publicly available NASA lidar data over the Maya region, identifying a previously unregistered archaeological site in Mexico within 72 hours. Lacking a specialized classifier, 48 agents visually inspected rendered lidar strips using a structured guide to detect geometric anomalies—straight edges, sharp corners, levelled plazas, and pyramidal mounds—while filtering natural features like sinkholes. The agents identified a hilltop complex featuring an oval pyramid 8–12 m tall, a raised courtyard, a flat-bottomed hollow (potential aguada), and three surrounding platforms, consistent with a Classic period (c. AD 250–900) civic center. After verifying the structures against raw point-cloud data (ground vs. vegetation returns), the agents drafted a Spanish-language report and submitted it to Mexico’s INAH. INAH confirmed no registered site existed at those coordinates, though subsequent literature review revealed the complex had been mapped in Estrada-Belli et al. (2025), meaning the agents rediscovered rather than discovered it. The workflow demonstrates that general-purpose agents can execute domain-specific visual analysis and bureaucratic navigation without fine-tuned models, leveraging existing remote-sensing datasets (cf. Auld-Thomas et al. 2024; Golden et al. 2016; Šprajc 2026) to accelerate archaeological survey.