A list of buildings can become a map-based screening task rather than a spreadsheet checked one address at a time. Google expanded experimental access to its Earth AI geospatial agent on October 14.
The agent combines Gemini with specialized models and tools for satellite imagery, weather, population information and mapping. Its purpose is to connect these sources through several analysis steps, rather than answer from a language model's memory alone.
Users can supply their own locations as spreadsheets, map boundaries or cloud tables. The agent plans the work, selects relevant tools and processes large map layers in a separate workspace. Its responses show the plan and data sources.
One partner example concerns potential community resilience hubs. According to Google, Milken Institute teams screened more than 1,500 candidate buildings using information including rooftop solar potential, nearby transit and population data.
That is a practical distinction: identifying buildings worth a closer look is different from certifying that a roof can support panels. The example describes screening with available datasets, not an engineering inspection of each site.
Google's evaluation compares different agent configurations across four benchmarks. These include versions with basic tools, Earth AI tools and an explicit planning stage. The researchers report that specialized tools and planning improve performance.
Access remains experimental and limited. Google offers an interest form, an expanded Ask Google Earth preview and an agent-to-agent service for developers connecting their own workflows.
The output includes the data trail behind the map.
