Back to all posts

AI Agents Analyze Maya Lidar Data Without a New Model

Forty-eight AI agents reviewed lidar images covering about 1,200 km² of the Maya region in less than 72 hours. The notable part is not a custom archaeology model—it’s the guided workflow around general-purpose agents, and the validation questions that remain.

AI agentsMaya regionlidar stripsraw point-cloud dataagentic workflow
main thumbnail for AI Agents Analyze Maya Lidar Data Without a New Model
main thumbnail for AI Agents Analyze Maya Lidar Data Without a New Model
Reader Lens

Automation needs a narrow first win

The best first AI workflow is usually a repeated task with a clear input, clear output, and a human approval step.

AI agents reviewed public lidar imagery across about 1,200 km² of the Maya region in less than 72 hours. The useful detail is not a new archaeological model: 48 agents followed a precise guide to inspect images, then helped check records and prepare a report for INAH. That makes this a workflow worth examining, not proof that agents can independently identify archaeological sites.

AI agents turned lidar strips into a review workflow

The project began with the author asking Gemini, ChatGPT, Claude, and Grok for ideas; the agents recommended lidar. The lidar strips were converted into images, which the agents examined one by one. The source says there was no special model or classifier involved: the agents looked at the pictures according to a guide.

That distinction changes what the example demonstrates. It is not evidence that a specialized vision system can detect Maya structures, nor that an agent can discover them without direction. It shows that an agentic workflow can apply general-purpose systems to a visual review task when the inputs are prepared and the inspection criteria are made explicit.

inside paper visual for AI Agents Analyze Maya Lidar Data Without a New Model
main thumbnail for AI Agents Analyze Maya Lidar Data Without a New Model

Worth a look

Phugialy Picks

AI Engineering: Building Applications with Foundation Models
Amazon

AI Engineering: Building Applications with Foundation Models

A practical guide to building real-world applications with foundation models and LLMs.

GMKtec K15 AI Mini PC Oculink Intel Ultra 5 125U 32GB DDR5 512GB SSD | Desktop Computer AI Boost, 3X M.2 2280 Storage Expansion, Dual NIC...
Amazon

GMKtec K15 AI Mini PC Oculink Intel Ultra 5 125U 32GB DDR5 512GB SSD | Desktop Computer AI Boost, 3X M.2 2280 Storage Expansion, Dual NIC...

2084 and the AI Revolution, Updated and Expanded Edition: How Artificial Intelligence Informs Our Future
Amazon

2084 and the AI Revolution, Updated and Expanded Edition: How Artificial Intelligence Informs Our Future

Some Phugialy Picks use affiliate links. If you buy through one, Phugialy may earn a commission. It doesn't change what we recommend. Full disclosure →

The useful work continued beyond image inspection

The agents also helped search literature and archives, check raw point-cloud data, and prepare a report for INAH. That makes the project more interesting than a batch of image descriptions. It links visual screening to follow-up work that could help someone assess candidate structures and assemble relevant records.

My read is that the practical unit here is the whole workflow, not the count of agents. Forty-eight agents and a large image set sound like scale, but the account does not establish how many candidate structures were found, how accurate the reviews were, or how much human checking was required. Without those details, speed and coverage are visible; reliability is not.

The next test is consistency, not another demo

For an integration, the hard question is what happens when images are ambiguous or the guide leaves room for judgment. A clear instruction can make a review repeatable, but it does not guarantee that different agents will interpret borderline terrain the same way. That is a production concern to test, not a failure established by this project.

I would treat this as a promising pattern for triage: use agents to cover a large visual collection, then make the review and verification steps explicit. Before relying on it for archaeological conclusions, I would want to see how candidate findings are checked against raw point-cloud data and records, and whether independent reviewers reach consistent results. The source shows agents contributing to those checks, but does not report an accuracy measure or a final validation outcome.

The broader point is that agentic AI is not confined to coding, math, and physics. This example makes a grounded case for trying it on document-and-image workflows, while leaving the key deployment question open: can the process produce consistent, verifiable findings when the material is messy?

Holiday picks

Thanksgiving: early shopping

Start early: products for hosting, travel and home, plus the gear you have been watching.

We earn from qualifying purchases made through these links, at no extra cost to you, as an Amazon Associate and through other affiliate programs. Disclosure. See all Thanksgiving picks

Source and trust note

Built from source research and filtered through practical implementation judgment.

Reference: pietromiotti.com

Got a question about how this applies to you? →

Keep reading

Follow the thread