Ask an AI assistant: “Find me a pharmacy that will still be open when I get there, with parking nearby, on my way home.”
It sounds straightforward. But answering it reliably requires more than understanding the question. The system must determine where you are, identify relevant pharmacies, account for current traffic, calculate whether they can be reached in time and factor in details such as parking and accessibility.
This illustrates a growing challenge as generative AI and agentic applications move beyond answering questions and begin interacting with the physical world.
Large language models (LLMs) are effective at interpreting intent and generating natural language responses. But they do not calculate valid routes through live road networks, verify access restrictions or determine whether a destination can be reached under current conditions.
The central challenge is not whether AI can describe places. It is whether AI can reason over changing physical networks with enough precision to support action.
India Makes the Location Challenge Interesting
Navigation is a computational problem. It requires calculating a valid path through a road graph and weighing constraints such as traffic, incidents, turn restrictions, access rules and vehicle type. A language model operates differently, generating responses based on patterns in data and language.
That distinction is particularly relevant in India. Addressing can be highly localised and landmark-based, places may have alternate names, and roads and infrastructure can change rapidly. Information such as commercial vehicle restrictions, parking availability and traffic conditions may sit across different data sources and change at different speeds.
The range of mobility use cases is also expanding. Passenger cars, two-wheelers, commercial vehicles, ride-hailing, food delivery, e-commerce, quick commerce and digitised supply chains each introduce different routing constraints, service-level expectations and operational costs when location decisions are wrong.
For these applications, knowing where something is represents only part of the problem.
Increasingly, we need to understand: What is the best way to get there right now? How long will it really take? Is this route appropriate for this vehicle? What conditions could affect the journey?
From Finding a Place to Reasoning About a Journey
This becomes even more important as vehicles become connected and electrified.
Consider an EV driver travelling from Mumbai to Pune. The question is no longer simply how to get from one city to another. The system may need to consider the vehicle’s available range, traffic conditions and charging options along the journey before determining whether a stop is needed and where it should be.
For a ride-hailing or delivery driver in a congested city, the system must identify the right pickup or delivery point, account for access and stopping constraints, update ETAs as conditions change and minimise avoidable delays that affect driver productivity and customer experience.
Generative AI can make these interactions far more intuitive. Instead of searching separately for a destination, parking, traffic conditions or another point of interest, a user can express a complex need conversationally. The system then needs to decompose that request, call the right location services and return an answer that reflects real-world conditions.
The challenge is turning that request into an answer that can actually be executed in the physical world. This is where the combination of AI and location intelligence becomes powerful.
Giving AI an Execution Layer for the Physical World
Adding geographic information to an LLM can help it recognise places and interpret spatial concepts more effectively. But this does not change the model’s fundamental operating logic: an LLM remains a system designed to generate probabilistic responses through language.
For AI to translate intent reliably into real-world action, a clearer division of labour is needed.
The LLM can infer what the user wants to achieve. A spatial execution layer can determine whether that goal is feasible and how to complete it. It can resolve places, calculate routes, process traffic information, apply road attributes and restrictions, evaluate alternatives and return a structured result. The language model can then explain the outcome naturally instead of attempting to generate the spatial logic itself.
With Location Reasoning, spatial intent is identified and handed to a purpose-built execution layer that performs the necessary location operations.
The Map Becomes More Important in the Age of AI, Not Less
This has particular relevance as AI starts playing a bigger role across mobility and enterprise applications in India.
Drivers increasingly expect real-time traffic and incident information that helps them make better decisions. EV drivers need greater confidence in planning their journeys. Logistics and delivery businesses need to improve efficiency across first-mile, middle-mile and last-mile operations. Across these use cases, safety and convenience remain fundamental expectations.
Generative AI could transform how people interact with these services. Instead of navigating menus, entering multiple searches or switching between applications, users will increasingly be able to express what they want conversationally and let AI determine which services are needed to fulfill that request.
The interface may become simpler, but what happens underneath it becomes more sophisticated.
As AI moves from providing information to helping people and businesses act in the physical world, spatial precision becomes more important, not less. Digital maps and location intelligence provide the reference layer for movement, accessibility, restrictions and changing conditions that AI needs to operate reliably in the real world.
LLMs can understand what we want to achieve. Location intelligence can help determine how to make it happen.
Abhijit Sengupta is General Manager, India and Southeast Asia at HERE Technologies. All views expressed are the author's own.