September 2026 Summaries

Felt and Databricks describe building a GIS-focused AI agent intended to let users ask spatial questions in natural language and receive live, interactive maps rather than static exports, code snippets, or unverified answers. In a demonstration using Databricks Agent Bricks and Felt’s MCP server, subagents identified relevant Arizona energy datasets, used spatial SQL through Lakebase’s PostGIS capabilities to find transmission infrastructure near active wildfires, and created a layered map of fire perimeters, lines, substations, burn probability, and solar assets in Slack. The architecture combines the Lakehouse for large-scale analytics with Lakebase for fast conversational geospatial queries, while Unity Catalog provides shared governance, permissions, lineage, and query traces. The approach is aimed at engineers, analysts, and platform teams that need governed mapping without building separate GIS workflows, although the authors emphasize that human expertise remains necessary to choose appropriate datasets, interpret results, and make decisions.

Felt has introduced integrated raster analysis capabilities to its collaborative GIS platform, enabling users to perform workflows previously requiring separate infrastructure and specialized tools. Available through the Felt AI Assistant and MCP, the new functions include zonal statistics, raster-band analysis, spatial filtering and clipping, elevation and route profiles, and enrichment of vector features with raster-derived values. Users can sample pixel values for points, lines, and polygons; create dashboard components such as histograms and filters from raster bands or derived indices including NDVI; and interactively update results when boundaries, routes, or selected ranges change. Felt positions its AI tools as a way to automate tasks such as locating relevant raster data, combining it with vector data, and ranking candidate areas, illustrated by a renewable-energy siting example involving wind conditions and existing turbines.