Large language models are famously bad at geometry. Ask one how far apart two addresses are and it will confidently guess. The two big location-platform vendors spent 2025 and 2026 turning that weakness into a product category — first by putting AI into navigation, and now by putting maps under AI agents.
HERE's answer: don't let the LLM do the math
The most interesting recent move is HERE's Location Reasoning, unveiled in May 2026. Rather than teaching a model geography, it offloads spatial computation from the LLM into "structured execution flows," promising "deterministic results: the same inputs and constraints produce the same answer every time without guessing or hallucinating" [1]. That is a direct architectural response to the grounding problem — the LLM orchestrates; the geospatial engine computes.
The data moat behind it is considerable: HERE maintains more than 68 million kilometers of roads across 200+ territories, with 238 million vehicles using its services [1]. At CES 2025 the company also announced a decade-long partnership with AWS to build AI-powered "live" maps [2].
Mapbox's answer: AI in the dashboard, tools for the builder
Mapbox came at it from the vehicle side. At CES 2025 it showed MapGPT-powered 3D lane-level guidance developed with Hyundai AutoEver, a Virtual Head Unit built with Arm, and a voice partnership with Cerence AI [3]. The Spring 2025 release then delivered the developer-facing substance: a Geofencing API in public preview, a Navigation UX Framework in beta, Zone Avoidance in beta, Boundaries v4.5 with 107,000+ new boundaries, and interactive map features in GL JS v3.11 [4].
What it means for GIS professionals
Two takeaways. First, "location intelligence" is being repackaged as an AI-grounding service — the buyers are increasingly agent builders, not cartographers. Second, determinism is becoming a selling point. After two years of generative demos, HERE marketing the absence of hallucination as a headline feature tells you where enterprise buyers got burned. For anyone whose job involves vetting spatial answers, that skepticism is a professional tailwind.