Whitmore
All insightsNewsModels & researchTools & workflows

Google’s Earth AI agents turn maps into answers

Google is expanding experimental access to agents that combine geographic data and specialist tools. Here is what changed and what remains in early access.

Written byWhitmore
Published15 Oct 2026
Reading time2 min read
CategoryModels & research
Google’s Earth AI public-health collage combines a health worker and patient with satellite imagery and map views.

A map can show you a building. Working out whether that building is a good place for a community resilience hub is a different job.

Google is expanding experimental access to an AI agent built for that second kind of question. Its October 14 update pairs a new research report with access through Ask Google Earth and a service developers can connect to their own workflows. Google Research

From a question to an analysis

The Geospatial Reasoning agent combines Gemini with specialist tools, geographic datasets and a plan for how to use them. The research compares that setup with simpler alternatives, including a language model alone and one equipped with basic tools. The authors report that the specialist tools make the biggest difference; planning adds value on open-ended tasks. Research paper

The practical idea is straightforward: turn a question into a sequence of geographic operations, then produce something a person can inspect.

Google’s official demonstrations show workflows involving uploaded location data, satellite imagery, routing and model evaluation. They make the ambition easier to picture: bring the relevant information together inside a working analysis, instead of asking someone to assemble every layer manually. Official demonstrations

A concrete use for it

Google says the Milken Institute tested the agent to screen candidate buildings for community resilience hubs, combining information such as rooftop solar potential, transit access and climate risks. The agent also exposes its analytical plan and data sources. Google Research

That is a useful example because the decision is about relationships between places. A building’s address tells you very little on its own. Its surroundings, access and constraints can change the picture.

Our take: the interesting shift is from asking where something is to asking which places meet a set of conditions, with a trail of work that someone can question.

Still an experiment

Access to the expanded research capabilities remains limited. Google is inviting organizations to express interest, rather than announcing unrestricted availability. Access details

For Ask Google Earth specifically, Google warns that responses can be inaccurate. It also says human reviewers may process queries and feedback, and tells users to keep confidential information out of those conversations. Ask Google Earth documentation

For a team exploring this, a sensible first question is one it already understands well enough to check. A useful map should make the reasoning easier to examine, not just make the answer look convincing.

Image: Google.

WhitmoreAI consulting and implementation

The reporting

Sources & context

  1. Putting Earth AI to work: Geospatial agents for our planet’s biggest challenges ↗

    Google Research · Read 15 Oct 2026

    October 14 announcement, partner example, disclosed planning and source information.

  2. Google Earth AI Geospatial Reasoning Agent: Capabilities and a Systematic Evaluation ↗

    Slonim et al. / arXiv · Read 15 Oct 2026

    Modular skills, upfront planning and vendor research evaluation; submitted October 11.

  3. Google Research Geospatial Reasoning ↗

    Google Research · Read 15 Oct 2026

    Product demonstrations and limited early-access routes.

  4. Overview of Ask Google Earth ↗

    Google Developers · Read 15 Oct 2026

    Ask Google Earth experimental status and query/data-review policy.

Let’s put AI to work in your business.

Tell us about your business. We will come back with what we would build, what we would skip, and what it takes to run it in production.